{ "dataset": "supply-chain", "version": "1.4.0", "generated_on": "2026-10-14", "licence": "CC BY 4.0 for ComputeFinder-owned summaries, labels and compilation rights only, except where a record supplies a specific adaptation licence. Third-party material retains its source terms.", "licence_url": "https://creativecommons.org/licenses/by/4.0/", "licence_scope": "Only copyright and database rights controlled by ComputeFinder in its original contributions. No third-party rights are granted. Source reuse terms have not been comprehensively cleared; consult each linked source. Private submissions, contact details, code and branding are excluded.", "source_terms_status": "Not comprehensively cleared; check linked sources before reusing third-party material.", "attribution": "Compiled by ComputeFinder (https://computefinder.net). Original contributions licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/), except adaptations carrying a record-specific licence. Give appropriate credit, retain licence and source notices, and indicate changes when sharing. Each record links its primary source; retain any attribution required by that source. Material is provided as-is without warranties.", "note": "No prices are included. Missing values are null or absent, never zero.", "records": [ { "id": "cf-sup-0001", "slug": "nvidia-blackwell", "name": "NVIDIA (Blackwell data centre GPUs)", "layer": "chips", "sublayer": "accelerators", "kind": "company", "hq_country": "US", "url": "https://www.nvidia.com/en-us/data-center/technologies/blackwell-architecture/", "summary": "NVIDIA designs the GPUs and rack-scale systems that many cloud and neocloud AI clusters are built around. Its Blackwell generation joins two large dies in one package and is sold as boards, racks and DGX systems.", "role": "Designer of the accelerator hardware you are most likely to see listed by GPU clouds, so its product names are the usual vocabulary for capacity listings.", "constraint": "Blackwell parts are built on a TSMC process (NVIDIA names 4NP) and depend on the foundry, packaging and memory layers below.", "facts": [ { "text": "NVIDIA states that Blackwell GPUs use two reticle-limited dies joined by a 10 TB/s chip-to-chip link and are made on TSMC 4NP.", "source_url": "https://www.nvidia.com/en-us/data-center/technologies/blackwell-architecture/" }, { "text": "NVIDIA lists GB200 NVL72 and GB300 NVL72 rack-scale systems, HGX B300 and DGX SuperPOD among its Blackwell-based products.", "source_url": "https://www.nvidia.com/en-us/data-center/technologies/blackwell-architecture/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/nvidia-blackwell", "permalink": "https://computefinder.net/entity/cf-sup-0001" }, { "id": "cf-sup-0002", "slug": "amd-instinct", "name": "AMD Instinct", "layer": "chips", "sublayer": "accelerators", "kind": "company", "hq_country": "US", "url": "https://www.amd.com/en/products/accelerators/instinct.html", "summary": "AMD's Instinct line is its data centre GPU accelerator family for AI and high-performance computing. The current pages list MI200, MI300, MI350 and MI400 series parts.", "role": "The main merchant alternative GPU family in cloud listings, relevant if you want a second source of accelerators or an open software stack.", "constraint": null, "facts": [ { "text": "AMD lists the MI455X and MI430X as its MI400 series, the latter aimed at sovereign AI and HPC.", "source_url": "https://www.amd.com/en/products/accelerators/instinct.html" }, { "text": "Instinct software is built on AMD's ROCm open software platform, alongside a Kubernetes-based Enterprise AI Reference Stack.", "source_url": "https://www.amd.com/en/products/accelerators/instinct.html" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/amd-instinct", "permalink": "https://computefinder.net/entity/cf-sup-0002" }, { "id": "cf-sup-0003", "slug": "intel-gaudi", "name": "Intel Gaudi", "layer": "chips", "sublayer": "accelerators", "kind": "company", "hq_country": "US", "url": "https://www.intel.com/content/www/us/en/products/details/processors/ai-accelerators/gaudi.html", "summary": "Gaudi 3 is Intel's data centre AI accelerator, offered as PCIe cards, mezzanine cards and baseboards. It uses standard Ethernet for scaling rather than a proprietary interconnect.", "role": "A smaller-volume accelerator option that appears in some OEM servers and a few clouds.", "constraint": null, "facts": [ { "text": "Intel's Gaudi page lists Gaudi 3 in PCIe, mezzanine and baseboard forms, sold through OEMs including Dell, HPE and Supermicro.", "source_url": "https://www.intel.com/content/www/us/en/products/details/processors/ai-accelerators/gaudi.html" }, { "text": "Intel names IBM Cloud and Denvr Dataworks among cloud routes to Gaudi.", "source_url": "https://www.intel.com/content/www/us/en/products/details/processors/ai-accelerators/gaudi.html" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/intel-gaudi", "permalink": "https://computefinder.net/entity/cf-sup-0003" }, { "id": "cf-sup-0004", "slug": "google-cloud-tpu", "name": "Google Cloud TPU", "layer": "chips", "sublayer": "accelerators", "kind": "company", "hq_country": "US", "url": "https://cloud.google.com/tpu", "summary": "TPUs are accelerators Google designed for machine learning and rents through Google Cloud. They are used through GKE, TPU slices and pods rather than bought as cards.", "role": "A non-GPU accelerator you can only reach by renting from Google Cloud, with PyTorch, JAX and TensorFlow support.", "constraint": null, "facts": [ { "text": "Google's TPU page names Trillium (sixth generation) and Ironwood (seventh generation) as generally available, with TPU 8t and 8i listed as coming soon.", "source_url": "https://cloud.google.com/tpu" }, { "text": "Access is via Google Kubernetes Engine, TPU slices of pods and Cluster Director.", "source_url": "https://cloud.google.com/tpu" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/google-cloud-tpu", "permalink": "https://computefinder.net/entity/cf-sup-0004" }, { "id": "cf-sup-0005", "slug": "aws-trainium", "name": "AWS Trainium", "layer": "chips", "sublayer": "accelerators", "kind": "company", "hq_country": "US", "url": "https://aws.amazon.com/ai/machine-learning/trainium/", "summary": "Trainium is Amazon's in-house AI chip family for training and inference, available only as AWS instances and UltraServers.", "role": "A cloud-exclusive accelerator, relevant if your workload can run on the AWS Neuron software stack.", "constraint": null, "facts": [ { "text": "AWS describes Trainium3 as having 144 GB of HBM3e per chip, and UltraServers of up to 144 chips.", "source_url": "https://aws.amazon.com/ai/machine-learning/trainium/" }, { "text": "AWS says Trainium works with PyTorch, vLLM and Hugging Face Transformers through its Neuron SDK.", "source_url": "https://aws.amazon.com/ai/machine-learning/trainium/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/aws-trainium", "permalink": "https://computefinder.net/entity/cf-sup-0005" }, { "id": "cf-sup-0006", "slug": "microsoft-maia", "name": "Microsoft Maia", "layer": "chips", "sublayer": "accelerators", "kind": "company", "hq_country": "US", "url": "https://azure.microsoft.com/en-us/blog/azure-maia-for-the-era-of-ai-from-silicon-to-software-to-systems/", "summary": "Maia is Microsoft's own AI accelerator for Azure. The page we fetched describes the first generation, Maia 100, announced in 2023.", "role": "An in-house accelerator used inside Azure, so it is not something to source separately. The fetched page covers only the first generation; later generations were not verified.", "constraint": null, "facts": [ { "text": "Microsoft describes Maia 100 as its first in-house AI accelerator, built on a 5nm process and designed for large cloud AI workloads such as Copilot.", "source_url": "https://azure.microsoft.com/en-us/blog/azure-maia-for-the-era-of-ai-from-silicon-to-software-to-systems/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/microsoft-maia", "permalink": "https://computefinder.net/entity/cf-sup-0006" }, { "id": "cf-sup-0007", "slug": "meta-mtia", "name": "Meta MTIA", "layer": "chips", "sublayer": "accelerators", "kind": "company", "hq_country": "US", "url": "https://ai.meta.com/blog/next-generation-meta-training-inference-accelerator-AI-MTIA/", "summary": "MTIA is Meta's custom accelerator, built for its own ranking and recommendation workloads. The page we fetched covers the second generation.", "role": "An internal accelerator not offered to outside customers, listed for completeness of the custom silicon picture.", "constraint": null, "facts": [ { "text": "Meta says its next-generation MTIA moves to TSMC 5nm from 7nm and runs production recommendation models in data centres.", "source_url": "https://ai.meta.com/blog/next-generation-meta-training-inference-accelerator-AI-MTIA/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/meta-mtia", "permalink": "https://computefinder.net/entity/cf-sup-0007" }, { "id": "cf-sup-0008", "slug": "cerebras-wafer-scale-engine", "name": "Cerebras", "layer": "chips", "sublayer": "accelerators", "kind": "company", "hq_country": "US", "url": "https://www.cerebras.ai/chip", "summary": "Cerebras builds AI processors on a single full wafer rather than cutting the wafer into many small chips. The processors are sold in CS-series systems.", "role": "An alternative accelerator architecture, accessed through Cerebras systems or its cloud.", "constraint": null, "facts": [ { "text": "Cerebras lists its WSE-3 Turbo as having 4 trillion transistors and 900,000 cores on a 46,225 mm2 die, used in the CS-4 system.", "source_url": "https://www.cerebras.ai/chip" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/cerebras-wafer-scale-engine", "permalink": "https://computefinder.net/entity/cf-sup-0008" }, { "id": "cf-sup-0009", "slug": "sambanova", "name": "SambaNova", "layer": "chips", "sublayer": "accelerators", "kind": "company", "hq_country": "US", "url": "https://sambanova.ai/", "summary": "SambaNova makes dataflow AI chips (RDUs) and sells them as a cloud service, a software stack and on-premises racks.", "role": "An inference-focused alternative to GPUs, reachable through its API cloud or as racks you host.", "constraint": null, "facts": [ { "text": "SambaNova lists SN40-16 and SN50 as its fourth and fifth generation RDU platforms.", "source_url": "https://sambanova.ai/" }, { "text": "It is offered as SambaCloud (OpenAI-compatible APIs), SambaStack and SambaRack.", "source_url": "https://sambanova.ai/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/sambanova", "permalink": "https://computefinder.net/entity/cf-sup-0009" }, { "id": "cf-sup-0010", "slug": "tenstorrent", "name": "Tenstorrent", "layer": "chips", "sublayer": "accelerators", "kind": "company", "hq_country": null, "url": "https://tenstorrent.com/", "summary": "Tenstorrent builds RISC-V based AI accelerators and sells them as cards, workstations and servers, with an open-source software stack and IP licensing.", "role": "An open-source-leaning option for teams that want to buy hardware directly rather than rent cloud capacity.", "constraint": null, "facts": [ { "text": "Tenstorrent offers Blackhole cards, TT-QuietBox workstations and Galaxy servers.", "source_url": "https://tenstorrent.com/" }, { "text": "Its TT-Forge compiler is open source and stated to be in public beta, supporting PyTorch, JAX and ONNX.", "source_url": "https://tenstorrent.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/tenstorrent", "permalink": "https://computefinder.net/entity/cf-sup-0010" }, { "id": "cf-sup-0011", "slug": "tsmc-foundry", "name": "TSMC", "layer": "chips", "sublayer": "foundry", "kind": "company", "hq_country": "TW", "url": "https://www.tsmc.com/english/dedicatedFoundry/technology/logic", "summary": "TSMC is a contract chip manufacturer that makes processors designed by other companies, including many of the AI accelerators above.", "role": "Sits between accelerator designers and finished chips, so its process and packaging schedule shapes what accelerators reach the market.", "constraint": "Many accelerators above are made on TSMC processes and packaged with its CoWoS technology, a shared dependency.", "facts": [ { "text": "TSMC states its N2 nanosheet process entered volume production in the fourth quarter of 2025.", "source_url": "https://www.tsmc.com/english/dedicatedFoundry/technology/logic" }, { "text": "TSMC lists A16, A14 and A12 as further logic nodes, A16 using backside power delivery.", "source_url": "https://www.tsmc.com/english/dedicatedFoundry/technology/logic" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/tsmc-foundry", "permalink": "https://computefinder.net/entity/cf-sup-0011" }, { "id": "cf-sup-0012", "slug": "samsung-foundry", "name": "Samsung Foundry", "layer": "chips", "sublayer": "foundry", "kind": "company", "hq_country": "KR", "url": "https://semiconductor.samsung.com/foundry/", "summary": "Samsung's contract manufacturing arm makes chips for outside designers and offers 3nm gate-all-around production and 2.5D and 3D packaging.", "role": "An alternative to TSMC for accelerator and custom silicon designers.", "constraint": null, "facts": [ { "text": "Samsung Foundry lists nodes from 28nm FD-SOI to 3nm GAA with EUV, plus 2.5D and 3D packaging and HPC/AI-oriented services.", "source_url": "https://semiconductor.samsung.com/foundry/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/samsung-foundry", "permalink": "https://computefinder.net/entity/cf-sup-0012" }, { "id": "cf-sup-0013", "slug": "intel-foundry", "name": "Intel Foundry", "layer": "chips", "sublayer": "foundry", "kind": "company", "hq_country": "US", "url": "https://www.intel.com/content/www/us/en/foundry/overview.html", "summary": "Intel Foundry offers manufacturing, advanced packaging and test to outside customers as well as to Intel's own products.", "role": "A US-based manufacturing and packaging route for designers wanting supply geography beyond Taiwan and Korea.", "constraint": null, "facts": [ { "text": "Intel describes 2D, 2.5D and 3D packaging and a full test suite as part of Intel Foundry services.", "source_url": "https://www.intel.com/content/www/us/en/foundry/overview.html" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/intel-foundry", "permalink": "https://computefinder.net/entity/cf-sup-0013" }, { "id": "cf-sup-0014", "slug": "smic", "name": "SMIC", "layer": "chips", "sublayer": "foundry", "kind": "company", "hq_country": "CN", "url": "https://www.smics.com/en/site/about_summary", "summary": "SMIC is a Shanghai-based contract manufacturer making chips on 8-inch and 12-inch wafers at sites in Shanghai, Beijing, Tianjin and Shenzhen.", "role": "The main China-based foundry, relevant to anyone tracking domestic Chinese accelerator supply.", "constraint": null, "facts": [ { "text": "SMIC says it operates fabs in Shanghai, Beijing, Tianjin and Shenzhen and has offices in the US, Europe, Japan and Taiwan.", "source_url": "https://www.smics.com/en/site/about_summary" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/smic", "permalink": "https://computefinder.net/entity/cf-sup-0014" }, { "id": "cf-sup-0015", "slug": "rapidus", "name": "Rapidus", "layer": "chips", "sublayer": "foundry", "kind": "company", "hq_country": "JP", "url": "https://www.rapidus.inc/en/", "summary": "Rapidus is a Japanese start-up foundry developing 2nm gate-all-around logic chips at Chitose, Hokkaido.", "role": "A prospective additional leading-edge foundry, not yet a source of volume supply.", "constraint": "Rapidus states mass production is scheduled to begin in 2027, so it is not yet a volume source.", "facts": [ { "text": "Rapidus says its pilot line has operated since April 2025 and mass production is scheduled for 2027.", "source_url": "https://www.rapidus.inc/en/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/rapidus", "permalink": "https://computefinder.net/entity/cf-sup-0015" }, { "id": "cf-sup-0016", "slug": "sk-hynix-hbm", "name": "SK hynix", "layer": "chips", "sublayer": "memory", "kind": "company", "hq_country": "KR", "url": "https://www.skhynix.com/", "summary": "SK hynix is a South Korean memory maker whose high-bandwidth memory (HBM) is stacked next to AI accelerators.", "role": "A leading HBM source for accelerators, so its supply feeds directly into how many GPUs can be finished.", "constraint": "Reported in April 2026: its CFO said customer demand for DRAM and NAND exceeded its supply capability.", "facts": [ { "text": "In April 2026 SK hynix's CFO was reported as saying demand was exceeding supply and that it would ramp HBM4 shipments to NVIDIA.", "source_url": "https://view.asiae.co.kr/en/article/2026042310570988187" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/sk-hynix-hbm", "permalink": "https://computefinder.net/entity/cf-sup-0016" }, { "id": "cf-sup-0017", "slug": "samsung-hbm", "name": "Samsung Electronics (HBM memory)", "layer": "chips", "sublayer": "memory", "kind": "company", "hq_country": "KR", "url": "https://semiconductor.samsung.com/dram/hbm/", "summary": "Samsung's memory business sells HBM3, HBM3E, HBM4 and HBM4E stacks for AI accelerators.", "role": "One of the three HBM suppliers, so a source of alternative memory supply for accelerator makers.", "constraint": null, "facts": [ { "text": "Samsung lists HBM4 at 36 GB per stack and HBM4E as up to 64 GB in a 16-high stack.", "source_url": "https://semiconductor.samsung.com/dram/hbm/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/samsung-hbm", "permalink": "https://computefinder.net/entity/cf-sup-0017" }, { "id": "cf-sup-0018", "slug": "micron-hbm", "name": "Micron (HBM memory)", "layer": "chips", "sublayer": "memory", "kind": "company", "hq_country": "US", "url": "https://www.micron.com/products/memory/hbm", "summary": "Micron is the US-based HBM supplier, offering HBM3E and HBM4 stacks for AI platforms.", "role": "The only US-headquartered HBM maker among the big three, useful for supply geography.", "constraint": "Reported in August 2025 that Micron's 2026 HBM supply was sold out, per TrendForce citing Micron's Chief Business Officer.", "facts": [ { "text": "Micron lists HBM4 at 36 GB in a 12-high stack and says it is in high-volume production.", "source_url": "https://www.micron.com/products/memory/hbm" }, { "text": "TrendForce reported in August 2025 that Micron said its 2026 HBM allocation was sold out.", "source_url": "https://www.trendforce.com/news/2025/08/13/news-hbm-battle-heats-up-micron-reportedly-hints-2026-sell-out-sk-hynix-yet-to-confirm" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/micron-hbm", "permalink": "https://computefinder.net/entity/cf-sup-0018" }, { "id": "cf-sup-0019", "slug": "tsmc-3dfabric-cowos", "name": "TSMC 3DFabric and CoWoS packaging", "layer": "chips", "sublayer": "packaging", "kind": "technology", "hq_country": "TW", "url": "https://3dfabric.tsmc.com/english/dedicatedFoundry/technology/3DFabric.htm", "summary": "CoWoS is TSMC's packaging technology for joining processor dies and memory on one carrier, part of its 3DFabric family alongside InFO and SoIC.", "role": "Packaging step that turns an accelerator die and its HBM stacks into a usable part.", "constraint": "See the advanced packaging bottleneck entry.", "facts": [ { "text": "TSMC describes CoWoS as its packaging option for high-performance computing, with InFO aimed at mobile and network uses and SoIC for vertical chiplet stacking.", "source_url": "https://3dfabric.tsmc.com/english/dedicatedFoundry/technology/3DFabric.htm" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/tsmc-3dfabric-cowos", "permalink": "https://computefinder.net/entity/cf-sup-0019" }, { "id": "cf-sup-0020", "slug": "ase-technology", "name": "ASE Technology", "layer": "chips", "sublayer": "packaging", "kind": "company", "hq_country": "TW", "url": "https://ase.aseglobal.com/", "summary": "ASE is an independent assembly and test provider (OSAT) offering 2.5D and 3D packaging, its FOCoS-Bridge technology and co-packaged optics.", "role": "An outsourced packaging route that sits beside TSMC's own packaging.", "constraint": null, "facts": [ { "text": "ASE promotes FOCoS-Bridge for AI and HPC packages and describes silicon photonics and co-packaged optics work.", "source_url": "https://ase.aseglobal.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/ase-technology", "permalink": "https://computefinder.net/entity/cf-sup-0020" }, { "id": "cf-sup-0021", "slug": "amkor-technology", "name": "Amkor Technology", "layer": "chips", "sublayer": "packaging", "kind": "company", "hq_country": "US", "url": "https://amkor.com/", "summary": "Amkor is a US-based outsourced packaging and test provider with sites across Asia, Europe and the US, and a new advanced packaging facility being built in Arizona.", "role": "A packaging alternative with US capacity under construction.", "constraint": null, "facts": [ { "text": "Amkor says it is building an advanced packaging and test facility in Arizona and has announced a partnership with NVIDIA.", "source_url": "https://amkor.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/amkor-technology", "permalink": "https://computefinder.net/entity/cf-sup-0021" }, { "id": "cf-sup-0022", "slug": "asml-euv", "name": "ASML", "layer": "chips", "sublayer": "equipment", "kind": "company", "hq_country": "NL", "url": "https://www.asml.com/en/products/euv-lithography-systems", "summary": "ASML makes the extreme ultraviolet lithography machines that print the finest chip layers, including a High-NA generation.", "role": "Equipment supplier behind leading-edge logic and DRAM manufacturing, upstream of every advanced accelerator.", "constraint": "ASML describes its EUV technology as unique to it, so leading-edge fabs depend on a single machine supplier.", "facts": [ { "text": "ASML states its EXE High-NA systems have a 0.55 numerical aperture against 0.33 for its NXE systems.", "source_url": "https://www.asml.com/en/products/euv-lithography-systems" }, { "text": "ASML reported 2025 total net sales of 32.7 billion euros.", "source_url": "https://investor.asml.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/asml-euv", "permalink": "https://computefinder.net/entity/cf-sup-0022" }, { "id": "cf-sup-0023", "slug": "applied-materials", "name": "Applied Materials", "layer": "chips", "sublayer": "equipment", "kind": "company", "hq_country": "US", "url": "https://www.appliedmaterials.com/us/en/semiconductor.html", "summary": "Applied Materials supplies deposition, material removal and measurement tools used across chip manufacturing and advanced packaging.", "role": "Equipment supplier for both logic and memory fabs, including HBM and packaging steps.", "constraint": null, "facts": [ { "text": "Applied Materials' semiconductor page highlights materials engineering, metrology and work on HBM and advanced packaging bottlenecks.", "source_url": "https://www.appliedmaterials.com/us/en/semiconductor.html" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/applied-materials", "permalink": "https://computefinder.net/entity/cf-sup-0023" }, { "id": "cf-sup-0024", "slug": "lam-research", "name": "Lam Research", "layer": "chips", "sublayer": "equipment", "kind": "company", "hq_country": "US", "url": "https://www.lamresearch.com/products/", "summary": "Lam Research makes etch, deposition and clean tools, with offerings for advanced packaging and cryogenic etch.", "role": "Equipment used in memory and logic fabs, including the 3D structures behind HBM.", "constraint": null, "facts": [ { "text": "Lam's product page lists etch, deposition, strip and clean, advanced packaging, cryogenic etch and metallization.", "source_url": "https://www.lamresearch.com/products/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/lam-research", "permalink": "https://computefinder.net/entity/cf-sup-0024" }, { "id": "cf-sup-0025", "slug": "tokyo-electron", "name": "Tokyo Electron", "layer": "chips", "sublayer": "equipment", "kind": "company", "hq_country": "JP", "url": "https://www.tel.com/product/", "summary": "Tokyo Electron makes coater/developers, etch, deposition and cleaning systems, including coater/developers supporting EUV and High-NA processes.", "role": "Works alongside lithography machines in the fab line.", "constraint": null, "facts": [ { "text": "Tokyo Electron lists coater/developer, plasma etch, deposition and wafer cleaning systems, with coater/developers supporting EUV including High-NA.", "source_url": "https://www.tel.com/product/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/tokyo-electron", "permalink": "https://computefinder.net/entity/cf-sup-0025" }, { "id": "cf-sup-0026", "slug": "kla", "name": "KLA", "layer": "chips", "sublayer": "equipment", "kind": "company", "hq_country": "US", "url": "https://www.kla.com/", "summary": "KLA makes inspection, metrology and process-control systems for chip, wafer, advanced packaging and substrate production.", "role": "Yield tooling that helps fabs and packagers find defects.", "constraint": null, "facts": [ { "text": "KLA describes process control and yield management across chip, wafer, advanced packaging and PCB or IC substrate manufacturing.", "source_url": "https://www.kla.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/kla", "permalink": "https://computefinder.net/entity/cf-sup-0026" }, { "id": "cf-sup-0027", "slug": "semi-equipment-market-data", "name": "SEMI equipment market data", "layer": "chips", "sublayer": "equipment", "kind": "programme", "hq_country": "US", "url": "https://www.semi.org/en/products-services/market-data/equipment", "summary": "SEMI is the industry body that publishes monthly and quarterly semiconductor equipment sales statistics and forecasts.", "role": "A neutral source for how much chip equipment is being bought, useful as context.", "constraint": null, "facts": [ { "text": "SEMI's page reports 2025 global equipment billings of 135 billion dollars, up 15 percent, and 23 percent year-on-year growth in the second quarter of 2026.", "source_url": "https://www.semi.org/en/products-services/market-data/equipment" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/semi-equipment-market-data", "permalink": "https://computefinder.net/entity/cf-sup-0027" }, { "id": "cf-sup-0028", "slug": "advanced-packaging-capacity", "name": "Advanced packaging capacity (CoWoS)", "layer": "chips", "sublayer": "packaging", "kind": "bottleneck", "hq_country": null, "url": "https://www.digitalcitizen.life/tsmc-reportedly-plans-to-double-cowos-capacity-by-2028-as-ai-packaging-demand-keeps-rising/", "summary": "Packaging dies and HBM together is a distinct manufacturing step, and trade press reports it has limited how many AI accelerators can be finished.", "role": "Limits how fast accelerator supply can grow, which affects availability of GPU capacity downstream.", "constraint": "Reported in September 2026 that advanced packaging is a limiting factor in AI chip production, with TSMC reportedly planning to roughly double CoWoS capacity by 2028.", "facts": [ { "text": "Digital Citizen reported on 12 September 2026 that TSMC planned to grow CoWoS capacity from about 130,000 to about 260,000 wafers a month by end of 2028 (reported, not confirmed by TSMC in the source).", "source_url": "https://www.digitalcitizen.life/tsmc-reportedly-plans-to-double-cowos-capacity-by-2028-as-ai-packaging-demand-keeps-rising/" }, { "text": "In July 2024 TSMC's chair said AI was driving strong demand for CoWoS and that TSMC was working with outsourced assembly and test firms to meet orders.", "source_url": "https://www.taiwannews.com.tw/en/news/5903663" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/advanced-packaging-capacity", "permalink": "https://computefinder.net/entity/cf-sup-0028" }, { "id": "cf-sup-0029", "slug": "hbm-supply", "name": "High-bandwidth memory supply", "layer": "chips", "sublayer": "memory", "kind": "bottleneck", "hq_country": null, "url": "https://view.asiae.co.kr/en/article/2026042310570988187", "summary": "HBM comes from only three suppliers, and trade press reports their output being committed ahead of time.", "role": "Each accelerator needs several HBM stacks, so memory supply caps accelerator volumes.", "constraint": "Reported in April 2026: SK hynix's CFO said customer demand exceeded supply capability and that it was hard to add supply quickly.", "facts": [ { "text": "Asiae reported on 23 April 2026 SK hynix's CFO saying demand exceeded supply and that suppliers found it difficult to raise supply in the short term.", "source_url": "https://view.asiae.co.kr/en/article/2026042310570988187" }, { "text": "TrendForce reported in August 2025 that Micron said its 2026 HBM supply was sold out.", "source_url": "https://www.trendforce.com/news/2025/08/13/news-hbm-battle-heats-up-micron-reportedly-hints-2026-sell-out-sk-hynix-yet-to-confirm" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/hbm-supply", "permalink": "https://computefinder.net/entity/cf-sup-0029" }, { "id": "cf-sup-0030", "slug": "glass-fibre-cloth-substrates", "name": "Glass fibre cloth for substrates", "layer": "chips", "sublayer": "materials", "kind": "bottleneck", "hq_country": null, "url": "https://insights.trendforce.com/p/glass-fiber-cloth-shortage", "summary": "Specialist glass fibre cloth goes into the substrates and circuit boards that carry AI chips, and TrendForce reports supply is tight.", "role": "A materials dependency below the packaging step that can delay accelerator and server output.", "constraint": "Reported by TrendForce in April 2026: a small number of suppliers dominate and capacity relief is not expected until about mid-2027.", "facts": [ { "text": "TrendForce reported on 30 April 2026 that Nittobo holds roughly 90 percent of T-glass supply and that constraints were not expected to ease until mid-2027 (analyst estimate, not company confirmed).", "source_url": "https://insights.trendforce.com/p/glass-fiber-cloth-shortage" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/glass-fibre-cloth-substrates", "permalink": "https://computefinder.net/entity/cf-sup-0030" }, { "id": "cf-sup-0031", "slug": "supermicro", "name": "Supermicro", "layer": "systems", "sublayer": "servers", "kind": "company", "hq_country": "US", "url": "https://ir.supermicro.com/", "summary": "San Jose server and rack maker that sells modular \"building block\" systems for AI, cloud, storage and edge. It sells both individual servers and integrated liquid-cooled racks.", "role": "A direct source of GPU servers and full racks for buyers and cloud operators who want configurable systems without going through a large OEM.", "constraint": "Its direct liquid cooling designs assume facilities with coolant distribution and warm-water loops, so rack delivery depends on the host data centre being ready (company press release).", "facts": [ { "text": "Supermicro describes its Building Block Solutions approach as the basis for customised, quickly delivered servers across AI, cloud, HPC and edge.", "source_url": "https://ir.supermicro.com/" }, { "text": "Its DLC-2 direct liquid cooling offer, announced on 14 May 2025, covers cold plates for CPUs, GPUs, memory, PCIe switches and voltage regulators, plus in-rack coolant distribution units.", "source_url": "https://www.supermicro.com/en/pressreleases/supermicros-dlc-2-next-generation-direct-liquid-cooling-solutions-aims-reduce-data" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/supermicro", "permalink": "https://computefinder.net/entity/cf-sup-0031" }, { "id": "cf-sup-0032", "slug": "dell-poweredge-ai-servers", "name": "Dell Technologies PowerEdge AI servers", "layer": "systems", "sublayer": "servers", "kind": "company", "hq_country": "US", "url": "https://www.dell.com/en-us/shop/ipovw/poweredge-xe9780", "summary": "Dell's PowerEdge XE line of GPU servers, from air-cooled eight-GPU HGX boxes to rack-scale liquid-cooled systems built around NVIDIA Blackwell parts.", "role": "One of the large OEM routes by which enterprises and cloud providers buy supported GPU servers and racks with vendor service.", "constraint": "The XE9712 rack-scale design is offered with direct liquid cooling to reach higher rack power density, so it needs a facility able to take liquid-cooled racks (Dell spec sheet).", "facts": [ { "text": "The PowerEdge XE9780 is an air-cooled server taking eight NVIDIA HGX B300 or HGX B200 GPUs, aimed at AI training and inference in conventional data halls.", "source_url": "https://www.dell.com/en-us/shop/ipovw/poweredge-xe9780" }, { "text": "The PowerEdge XE9712 spec sheet describes 1U sleds with four Blackwell Ultra GPUs each, up to 72 GPUs in a 48U rack joined by NVLink, with air or direct liquid cooling options.", "source_url": "https://www.delltechnologies.com/asset/en-au/products/servers/technical-support/poweredge-xe9712-spec-sheet.pdf" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/dell-poweredge-ai-servers", "permalink": "https://computefinder.net/entity/cf-sup-0032" }, { "id": "cf-sup-0033", "slug": "hpe-cray-supercomputing-gx5000", "name": "HPE Cray Supercomputing GX5000", "layer": "systems", "sublayer": "servers", "kind": "company", "hq_country": "US", "url": "https://www.hpe.com/us/en/newsroom/blog-post/2026/07/hpe-delivers-industrys-highest-density-performance-for-ai-driven-science-with-supercomputing-powered-by-6th-gen-amd-epyc-processors.html", "summary": "Hewlett Packard Enterprise's liquid-cooled supercomputing platform for AI and science, built on 6th generation AMD EPYC processors and AMD Instinct accelerators.", "role": "Relevant to national labs and research centres that procure large HPC and AI systems rather than renting cloud instances.", "constraint": null, "facts": [ { "text": "HPE describes two GX5000 blades: an accelerated blade pairing one EPYC processor with four AMD Instinct MI430X GPUs (up to 112 GPUs per rack) and a CPU-only blade.", "source_url": "https://www.hpe.com/us/en/newsroom/blog-post/2026/07/hpe-delivers-industrys-highest-density-performance-for-ai-driven-science-with-supercomputing-powered-by-6th-gen-amd-epyc-processors.html" }, { "text": "HPE names Oak Ridge National Laboratory and Germany's HLRS as early adopters and describes the platform as fully direct-liquid-cooled.", "source_url": "https://www.hpe.com/us/en/newsroom/blog-post/2026/07/hpe-delivers-industrys-highest-density-performance-for-ai-driven-science-with-supercomputing-powered-by-6th-gen-amd-epyc-processors.html" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/hpe-cray-supercomputing-gx5000", "permalink": "https://computefinder.net/entity/cf-sup-0033" }, { "id": "cf-sup-0034", "slug": "lenovo-neptune", "name": "Lenovo Neptune liquid cooling and ThinkSystem AI servers", "layer": "systems", "sublayer": "servers", "kind": "technology", "hq_country": null, "url": "https://www.lenovo.com/us/en/servers-storage/neptune/", "summary": "Lenovo's family of liquid cooling options for its ThinkSystem servers, ranging from full direct water cooling to liquid-assisted air cooling and rear-door heat exchangers.", "role": "Gives buyers a vendor-supported route to dense GPU servers where air cooling is not enough, including systems that scale from single nodes to supercomputers.", "constraint": null, "facts": [ { "text": "Lenovo lists Neptune variants for full-system direct water cooling, targeted cooling of CPUs, GPUs and memory, liquid-assisted air cooling and rack-level heat exchangers.", "source_url": "https://www.lenovo.com/us/en/servers-storage/neptune/" }, { "text": "Neptune-equipped systems named on the page include the ThinkSystem SR780a V3 (eight NVIDIA GPUs), SR675 V3 and SC777 V4.", "source_url": "https://www.lenovo.com/us/en/servers-storage/neptune/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/lenovo-neptune", "permalink": "https://computefinder.net/entity/cf-sup-0034" }, { "id": "cf-sup-0035", "slug": "foxconn-hon-hai", "name": "Foxconn (Hon Hai Technology Group)", "layer": "systems", "sublayer": "servers", "kind": "company", "hq_country": "TW", "url": "https://aap.com.au/aapreleases/cision20250318ae43739/", "summary": "Taiwanese electronics manufacturer that builds AI servers and racks for the major accelerator vendors, with a liquid cooling arm through its Ingrasys subsidiary.", "role": "A contract manufacturer that sits between chip designers and the systems that cloud and AI operators eventually buy, so its output shapes how quickly rack-scale systems reach customers.", "constraint": null, "facts": [ { "text": "In a March 2025 release Foxconn said it works with NVIDIA on GB300 NVL72 racks and, as an NVIDIA pilot build partner, manufactures both GB200 NVL72 and GB300 NVL72 platforms.", "source_url": "https://aap.com.au/aapreleases/cision20250318ae43739/" }, { "text": "The same release credits its subsidiary Ingrasys with liquid cooling solutions designed using digital twin simulation.", "source_url": "https://aap.com.au/aapreleases/cision20250318ae43739/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/foxconn-hon-hai", "permalink": "https://computefinder.net/entity/cf-sup-0035" }, { "id": "cf-sup-0036", "slug": "wiwynn", "name": "Wiwynn", "layer": "systems", "sublayer": "servers", "kind": "company", "hq_country": "TW", "url": "https://wiwynn.com/about-wiwynn/company-info", "summary": "Taiwanese cloud infrastructure maker founded in 2012 as a subsidiary of Wistron, supplying servers, storage and integrated racks to hyperscale data centres.", "role": "An ODM that builds custom and OCP-based racks for large cloud operators, which is how much hyperscale capacity is actually assembled.", "constraint": null, "facts": [ { "text": "Wiwynn says it offers rack integration up to L11 level and works on a direct ODM model with custom, OCP-compliant designs.", "source_url": "https://wiwynn.com/about-wiwynn/company-info" }, { "text": "It reports platinum membership of the Open Compute Project, and its site lists AI infrastructure and advanced cooling among its offerings.", "source_url": "https://wiwynn.com/about-wiwynn/company-info" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/wiwynn", "permalink": "https://computefinder.net/entity/cf-sup-0036" }, { "id": "cf-sup-0037", "slug": "quanta-qct", "name": "Quanta Cloud Technology (QCT)", "layer": "systems", "sublayer": "servers", "kind": "company", "hq_country": "TW", "url": "https://www.qct.io/Company-AboutQCT", "summary": "Direct-sales data centre arm of Quanta Computer, launched in May 2012, selling servers, storage, switches and integrated racks to cloud providers, telecoms and enterprises.", "role": "A hyperscale-oriented hardware supplier whose open commodity designs are one route for operators building their own capacity.", "constraint": null, "facts": [ { "text": "Quanta set up its data centre business unit in 2000 and launched QCT as a direct-sales subsidiary in May 2012.", "source_url": "https://www.qct.io/Company-AboutQCT" }, { "text": "QCT says it designs, makes and services servers, storage, switches and integrated racks built on standard and open commodity hardware.", "source_url": "https://www.qct.io/Company-AboutQCT" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/quanta-qct", "permalink": "https://computefinder.net/entity/cf-sup-0037" }, { "id": "cf-sup-0038", "slug": "inspur-information", "name": "Inspur Information", "layer": "systems", "sublayer": "servers", "kind": "company", "hq_country": "CN", "url": "https://www.hpcwire.com/off-the-wire/inspur-launches-full-stack-liquid-cooled-server-solutions/", "summary": "Chinese server maker with a range from general-purpose and rack servers to GPU servers, including cold-plate liquid-cooled models. The source read is a July 2022 company release, so it shows the range at that date only.", "role": "A major server supplier in China, relevant to anyone mapping non-US sources of AI server hardware.", "constraint": null, "facts": [ { "text": "A July 2022 company release described cold-plate liquid cooling applied across its portfolio, including AI servers NF5688LM6 and NF5498LA5.", "source_url": "https://www.hpcwire.com/off-the-wire/inspur-launches-full-stack-liquid-cooled-server-solutions/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/inspur-information", "permalink": "https://computefinder.net/entity/cf-sup-0038" }, { "id": "cf-sup-0039", "slug": "nvidia-gb300-nvl72", "name": "NVIDIA GB300 NVL72", "layer": "systems", "sublayer": "servers", "kind": "technology", "hq_country": "US", "url": "https://www.nvidia.com/en-us/data-center/gb300-nvl72/", "summary": "A fully liquid-cooled rack-scale system that joins 72 Blackwell Ultra GPUs and 36 Grace CPUs into one NVLink domain.", "role": "The reference rack-scale unit that many cloud and neocloud GPU offerings are built from, so its availability and the racks that house it set what can be rented.", "constraint": "It is sold as a liquid-cooled rack, so deployment depends on facilities with liquid cooling and high rack power (NVIDIA product page).", "facts": [ { "text": "NVIDIA states the system combines 72 Blackwell Ultra GPUs and 36 Grace CPUs, with 130 TB/s of NVLink bandwidth across the GPU domain.", "source_url": "https://www.nvidia.com/en-us/data-center/gb300-nvl72/" }, { "text": "Each GPU gets 800 Gb/s of network connectivity through ConnectX-8 SuperNICs, usable with either Quantum-X800 InfiniBand or Spectrum-X Ethernet.", "source_url": "https://www.nvidia.com/en-us/data-center/gb300-nvl72/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/nvidia-gb300-nvl72", "permalink": "https://computefinder.net/entity/cf-sup-0039" }, { "id": "cf-sup-0040", "slug": "amd-helios-rack", "name": "AMD Helios rack-scale platform", "layer": "systems", "sublayer": "servers", "kind": "technology", "hq_country": "US", "url": "https://rocm.blogs.amd.com/ecosystems-and-partners/cdna5-helios/README.html", "summary": "AMD's open rack-scale design combining 72 Instinct MI455X GPUs, networking and management, built on the Open Rack Wide standard.", "role": "A second rack-scale route for AI capacity, using open scale-up and scale-out networking rather than a proprietary fabric.", "constraint": "Partner systems depend on AMD delivering the MI455X and new CPUs; one trade-press report on HPE's Helios offer said availability was pending that delivery.", "facts": [ { "text": "AMD's August 2026 blog describes 72 MI455X GPUs per rack with 31 TB of HBM4 memory, with scale-up using UALink over Ethernet and scale-out using Ultra Ethernet-based networking.", "source_url": "https://rocm.blogs.amd.com/ecosystems-and-partners/cdna5-helios/README.html" }, { "text": "It is reported that HPE will offer a Helios rack with a Juniper scale-up switch built on Broadcom Tomahawk 6 silicon.", "source_url": "https://www.theregister.com/2025/12/02/hpe_amd_helios_racks/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/amd-helios-rack", "permalink": "https://computefinder.net/entity/cf-sup-0040" }, { "id": "cf-sup-0041", "slug": "nvidia-infiniband-quantum", "name": "NVIDIA Quantum InfiniBand", "layer": "systems", "sublayer": "networking", "kind": "technology", "hq_country": "US", "url": "https://www.nvidia.com/en-us/networking/products/infiniband/", "summary": "NVIDIA's InfiniBand switching platform for supercomputers and AI clusters, including the 800G-class Quantum-X800 and in-network computing through SHARP.", "role": "A common back-end fabric in large training clusters, so it affects which interconnect a rented cluster uses.", "constraint": null, "facts": [ { "text": "NVIDIA describes SHARP as offloading collective operations to the switch network, reducing the data that crosses the fabric.", "source_url": "https://www.nvidia.com/en-us/networking/products/infiniband/" }, { "text": "Quantum-X800 is presented as the high-bandwidth InfiniBand switch for very large AI clusters.", "source_url": "https://www.nvidia.com/en-us/networking/products/infiniband/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/nvidia-infiniband-quantum", "permalink": "https://computefinder.net/entity/cf-sup-0041" }, { "id": "cf-sup-0042", "slug": "nvidia-spectrum-x", "name": "NVIDIA Spectrum-X Ethernet", "layer": "systems", "sublayer": "networking", "kind": "technology", "hq_country": "US", "url": "https://www.nvidia.com/en-us/networking/spectrumx/", "summary": "NVIDIA's Ethernet platform for AI, pairing Spectrum switches with SuperNICs and BlueField DPUs to run RDMA over Converged Ethernet between GPU servers.", "role": "The Ethernet alternative to InfiniBand inside NVIDIA-based clusters, and the networking used by many multi-tenant AI clouds.", "constraint": null, "facts": [ { "text": "NVIDIA lists Spectrum SN5000 and SN6000 series switches and SuperNICs as the core of the platform, aimed at multi-tenant AI factories.", "source_url": "https://www.nvidia.com/en-us/networking/spectrumx/" }, { "text": "The platform page also covers Spectrum-X Ethernet Photonics, which uses silicon photonics in place of pluggable transceivers, with power and resilience claims made by NVIDIA.", "source_url": "https://www.nvidia.com/en-us/networking/spectrumx/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/nvidia-spectrum-x", "permalink": "https://computefinder.net/entity/cf-sup-0042" }, { "id": "cf-sup-0043", "slug": "broadcom-tomahawk-6", "name": "Broadcom Tomahawk 6 and Tomahawk 6 Davisson", "layer": "systems", "sublayer": "networking", "kind": "technology", "hq_country": "US", "url": "https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-tomahawkr-6-davisson-industrys-first-1024", "summary": "Broadcom's 102.4 Tbps Ethernet switch silicon for AI clusters, with a Davisson variant that integrates optical engines in the package.", "role": "Merchant switch silicon used by many switch vendors, so it is a shared dependency beneath several branded AI Ethernet products.", "constraint": "Davisson was described at announcement (8 October 2025) as sampling to early access customers, so volume availability should be checked with vendors.", "facts": [ { "text": "Broadcom announced Tomahawk 6 Davisson on 8 October 2025 as its third-generation co-packaged optics switch with 102.4 Tbps of capacity and 200 Gbps per channel.", "source_url": "https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-tomahawkr-6-davisson-industrys-first-1024" }, { "text": "Broadcom states the design cuts optical interconnect power by 70 per cent against pluggable optics and targets two-tier networks above 100,000 accelerators.", "source_url": "https://investors.broadcom.com/news-releases/news-release-details/broadcom-announces-tomahawkr-6-davisson-industrys-first-1024" }, { "text": "Broadcom's product page positions Tomahawk 6 for both scale-up and scale-out AI networks.", "source_url": "https://www.broadcom.com/products/ethernet-connectivity/switching/strataxgs/bcm78910-series" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/broadcom-tomahawk-6", "permalink": "https://computefinder.net/entity/cf-sup-0043" }, { "id": "cf-sup-0044", "slug": "arista-networks", "name": "Arista Networks", "layer": "systems", "sublayer": "networking", "kind": "company", "hq_country": "US", "url": "https://www.arista.com/en/company/company-overview", "summary": "Santa Clara cloud networking company founded in 2008, selling Ethernet switches and its EOS operating system for AI and cloud data centres.", "role": "A supplier of Ethernet switching for AI back-end and front-end networks, relevant to operators choosing an Ethernet-based cluster fabric.", "constraint": null, "facts": [ { "text": "Arista lists the 7800R4 series at 800G for accelerated computing, and 7060 series platforms at 800G and 1600G for AI architectures.", "source_url": "https://www.arista.com/en/company/company-overview" }, { "text": "Arista states it is a founding member of the Ultra Ethernet Consortium.", "source_url": "https://www.arista.com/en/company/company-overview" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/arista-networks", "permalink": "https://computefinder.net/entity/cf-sup-0044" }, { "id": "cf-sup-0045", "slug": "cisco-nexus-silicon-one", "name": "Cisco Nexus 9000 and Silicon One for AI", "layer": "systems", "sublayer": "networking", "kind": "company", "hq_country": "US", "url": "https://www.cisco.com/c/en/us/products/collateral/networking/cloud-networking-switches/nexus-9000-switches/nexus-9000-ai-networking-wp.html", "summary": "Cisco's Ethernet AI networking line, pairing its own Silicon One chips with Nexus 9000 switches, NX-OS and Nexus Dashboard.", "role": "An enterprise-oriented source of AI cluster networking, offering a vendor-integrated alternative to NVIDIA fabrics.", "constraint": null, "facts": [ { "text": "Cisco's white paper describes the Nexus 9364E-SG2 as a 64-port 800G switch built on Silicon One G200, with a shared packet buffer for bursty AI traffic.", "source_url": "https://www.cisco.com/c/en/us/products/collateral/networking/cloud-networking-switches/nexus-9000-switches/nexus-9000-ai-networking-wp.html" }, { "text": "Cisco states it supports the Ultra Ethernet Consortium standard and uses features such as priority flow control and adaptive load balancing.", "source_url": "https://www.cisco.com/c/en/us/products/collateral/networking/cloud-networking-switches/nexus-9000-switches/nexus-9000-ai-networking-wp.html" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/cisco-nexus-silicon-one", "permalink": "https://computefinder.net/entity/cf-sup-0045" }, { "id": "cf-sup-0046", "slug": "hpe-juniper-networking", "name": "HPE Networking (including Juniper)", "layer": "systems", "sublayer": "networking", "kind": "company", "hq_country": "US", "url": "https://www.hpe.com/us/en/newsroom/press-release/2025/07/hewlett-packard-enterprise-closes-acquisition-of-juniper-networks-to-offer-industry-leading-comprehensive-cloud-native-ai-driven-portfolio.html", "summary": "HPE's networking business after it closed its purchase of Juniper Networks, combining campus, data centre and service provider routing and switching.", "role": "Consolidates a major data centre switching vendor inside a server and supercomputer supplier, so network and compute can be bought together.", "constraint": null, "facts": [ { "text": "HPE announced that the Juniper acquisition closed on 2 July 2025, with Juniper's former chief executive Rami Rahim leading HPE Networking.", "source_url": "https://www.hpe.com/us/en/newsroom/press-release/2025/07/hewlett-packard-enterprise-closes-acquisition-of-juniper-networks-to-offer-industry-leading-comprehensive-cloud-native-ai-driven-portfolio.html" }, { "text": "Juniper contributes data centre switching (QFX), routing (PTX) and security (SRX) product lines.", "source_url": "https://www.hpe.com/us/en/newsroom/press-release/2025/07/hewlett-packard-enterprise-closes-acquisition-of-juniper-networks-to-offer-industry-leading-comprehensive-cloud-native-ai-driven-portfolio.html" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/hpe-juniper-networking", "permalink": "https://computefinder.net/entity/cf-sup-0046" }, { "id": "cf-sup-0047", "slug": "ultra-ethernet-consortium", "name": "Ultra Ethernet Consortium", "layer": "systems", "sublayer": "standards", "kind": "programme", "hq_country": null, "url": "https://ultraethernet.org/", "summary": "A Linux Foundation-affiliated Joint Development Foundation project defining an open, Ethernet-based communications stack for AI and HPC at scale.", "role": "Sets the open transport standard that several switch and NIC vendors say they support, which can reduce lock-in to a single networking vendor.", "constraint": null, "facts": [ { "text": "The consortium's site lists specification version 1.0.3, published in August 2026, with release notes dated July 2026.", "source_url": "https://ultraethernet.org/" }, { "text": "Its stated scope covers multipathing, congestion management and tail latency, while keeping compatibility with existing Ethernet APIs.", "source_url": "https://ultraethernet.org/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/ultra-ethernet-consortium", "permalink": "https://computefinder.net/entity/cf-sup-0047" }, { "id": "cf-sup-0048", "slug": "ualink-consortium", "name": "UALink Consortium", "layer": "systems", "sublayer": "standards", "kind": "programme", "hq_country": null, "url": "https://ualinkconsortium.org/", "summary": "An industry group defining an open accelerator-to-accelerator interconnect for AI pods, as an alternative to proprietary scale-up links.", "role": "Defines the scale-up fabric inside a pod, which determines how many accelerators can act as one system in non-NVIDIA designs.", "constraint": null, "facts": [ { "text": "The 200G 1.0 specification was released on 8 April 2025 and supports up to 1,024 accelerators in a pod at 200G per lane.", "source_url": "https://www.businesswire.com/news/home/20250408050548/en/UALink-Consortium-Releases-the-Ultra-Accelerator-Link-200G-1.0-Specification" }, { "text": "Board members named in the release include AMD, AWS, Cisco, Google, HPE, Intel, Meta and Microsoft.", "source_url": "https://www.businesswire.com/news/home/20250408050548/en/UALink-Consortium-Releases-the-Ultra-Accelerator-Link-200G-1.0-Specification" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/ualink-consortium", "permalink": "https://computefinder.net/entity/cf-sup-0048" }, { "id": "cf-sup-0049", "slug": "open-compute-project", "name": "Open Compute Project", "layer": "systems", "sublayer": "standards", "kind": "programme", "hq_country": null, "url": "https://www.opencompute.org/about", "summary": "A community that shares open hardware designs for servers, racks, networking, storage and power, started by Facebook (now Meta) in 2011.", "role": "Its rack and power specifications underpin many hyperscale and ODM-built AI systems, including AMD Helios.", "constraint": null, "facts": [ { "text": "OCP was established in 2011 by Facebook and runs strategic workstreams called Open Systems for AI and Open Cluster Designs for AI.", "source_url": "https://www.opencompute.org/about" }, { "text": "On 14 October 2025 OCP launched an Open Data Center for AI initiative, started by Google, Meta and Microsoft, covering power, cooling, mechanical design and telemetry.", "source_url": "https://tools.prnewswire.com/en-us/live/20823/release/20251013EN96620" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/open-compute-project", "permalink": "https://computefinder.net/entity/cf-sup-0049" }, { "id": "cf-sup-0050", "slug": "800-vdc-power-architecture", "name": "800 VDC data centre power architecture", "layer": "systems", "sublayer": "standards", "kind": "technology", "hq_country": null, "url": "https://www.opencompute.org/documents/dcf-power-distribution-lvdc-white-paper-version-1-1-0-pdf", "summary": "800 VDC describes a direct-current distribution voltage, not the power of a rack or the capacity of a site.", "role": "OCP examines higher-voltage DC distribution as one approach to the electrical limits of high-density racks.", "constraint": "The architectural framework is not a dated register of installed sites. Adoption counts, named-site delivery and compute availability remain unverified here.", "facts": [ { "text": "OCP's LVDC white paper version 1.1.0 is dated 2 September 2026. Its introduction describes DC distribution as being evaluated and in early deployment, without establishing how many sites use 800 VDC.", "source_url": "https://www.opencompute.org/documents/dcf-power-distribution-lvdc-white-paper-version-1-1-0-pdf" } ], "questions": [ { "question": "What does 800 VDC mean?", "answer": "It means 800 volts of direct current. Voltage is not rack power, site capacity or an offer of compute access." }, { "question": "Does 800 VDC prove a 1 MW rack?", "answer": "No. The paper discusses higher voltage as a way towards high rack-power potential. A voltage label alone does not establish a rack's rated load, installed equipment or delivered capability." }, { "question": "Must every existing facility be rebuilt?", "answer": "There is no blanket conclusion. OCP expects AC and DC approaches to coexist. Compatibility depends on the site design, applicable codes and standards, equipment certification and approval by the responsible authority." }, { "question": "Is it deployed or still a design?", "answer": "The paper covers architectural choices, evaluation and early deployment. It is not proof that a particular site's 800 VDC system is commissioned. Ask for dated, phase-specific delivery evidence; neither universal adoption nor universal absence follows." } ], "source_notice": { "title": "OCP DCF Power Distribution LVDC white paper, version 1.1.0", "published_on": "2026-09-02", "source_url": "https://www.opencompute.org/documents/dcf-power-distribution-lvdc-white-paper-version-1-1-0-pdf", "attribution": "Adapted by ComputeFinder from the Open Compute Project and the contributors credited on pages 2 to 6 of the white paper.", "licence": "CC BY-SA 4.0", "licence_url": "https://creativecommons.org/licenses/by-sa/4.0/", "scope": "This record's summary, role, constraint, facts and question answers are a CC BY-SA 4.0 adaptation. The general CC BY 4.0 compilation notice does not override this scope.", "changes": "Shortened and rewritten as reader questions; no diagrams, tables, specifications or contributor contact details reproduced.", "disclaimer": "Provided as-is without warranties. No OCP or contributor endorsement. Patent and trademark rights are not granted. This is an explainer, not installation or electrical-safety advice." }, "checked_on": "2026-10-14", "page_url": "https://computefinder.net/supply-chain/800-vdc-power-architecture", "permalink": "https://computefinder.net/entity/cf-sup-0050" }, { "id": "cf-sup-0051", "slug": "nvidia-photonics-cpo-switches", "name": "NVIDIA Spectrum-X Photonics and Quantum-X Photonics", "layer": "systems", "sublayer": "optics", "kind": "technology", "hq_country": "US", "url": "https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-Announces-Spectrum-X-Photonics-Co-Packaged-Optics-Networking-Switches-to-Scale-AI-Factories-to-Millions-of-GPUs/", "summary": "NVIDIA's co-packaged optics switches, one Ethernet and one InfiniBand, which put optical engines on the switch package instead of using pluggable transceivers.", "role": "Shows how a major vendor is changing the optical layer of cluster networks, which affects transceiver and laser demand.", "constraint": "NVIDIA's announcement names a supplier group including TSMC, Coherent, Corning, Fabrinet, Foxconn, Lumentum and Sumitomo Electric, so these switches depend on several specialist optical suppliers (company release, March 2025).", "facts": [ { "text": "The March 2025 announcement described Quantum-X Photonics with 144 ports at 800 Gb/s and Spectrum-X Photonics configurations up to 400 Tb/s total bandwidth.", "source_url": "https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-Announces-Spectrum-X-Photonics-Co-Packaged-Optics-Networking-Switches-to-Scale-AI-Factories-to-Millions-of-GPUs/" }, { "text": "NVIDIA claims 3.5 times better power efficiency than its previous approach; the figure is the company's own.", "source_url": "https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-Announces-Spectrum-X-Photonics-Co-Packaged-Optics-Networking-Switches-to-Scale-AI-Factories-to-Millions-of-GPUs/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/nvidia-photonics-cpo-switches", "permalink": "https://computefinder.net/entity/cf-sup-0051" }, { "id": "cf-sup-0052", "slug": "marvell-co-packaged-optics", "name": "Marvell co-packaged optics for custom accelerators", "layer": "systems", "sublayer": "optics", "kind": "technology", "hq_country": "US", "url": "https://investor.marvell.com/news-events/press-releases/detail/114/marvell-announces-breakthrough-co-packaged-optics-architecture-for-custom-ai-accelerators", "summary": "An architecture Marvell announced in January 2025 for integrating optical engines into custom AI accelerator packages.", "role": "Targets cloud providers designing their own accelerators who want optical links between chips across racks rather than copper inside one rack.", "constraint": null, "facts": [ { "text": "Marvell said the approach would let operators move from tens of accelerators in a copper-linked rack to hundreds across racks.", "source_url": "https://investor.marvell.com/news-events/press-releases/detail/114/marvell-announces-breakthrough-co-packaged-optics-architecture-for-custom-ai-accelerators" }, { "text": "It cited a 6.4T 3D silicon photonics engine supporting 200 Gbps interfaces.", "source_url": "https://investor.marvell.com/news-events/press-releases/detail/114/marvell-announces-breakthrough-co-packaged-optics-architecture-for-custom-ai-accelerators" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/marvell-co-packaged-optics", "permalink": "https://computefinder.net/entity/cf-sup-0052" }, { "id": "cf-sup-0053", "slug": "coherent", "name": "Coherent", "layer": "systems", "sublayer": "optics", "kind": "company", "hq_country": "US", "url": "https://www.coherent.com/", "summary": "Photonics company supplying datacom transceivers, lasers and optical circuit switches for AI data centres, with in-house indium phosphide laser production.", "role": "One of the main makers of 800G and 1.6T optical modules that connect GPU clusters.", "constraint": "An analyst summary of Coherent's fiscal fourth quarter 2026 results reports that indium phosphide laser capacity, not assembly or test, is the limit on its output.", "facts": [ { "text": "Coherent describes itself as vertically integrated in datacom transceivers and lists a silicon photonics 1.6T-DR8 module and an optical circuit switch among its AI products.", "source_url": "https://www.coherent.com/" }, { "text": "It is reported that the company is moving indium phosphide production to six-inch wafers in Texas and Sweden, with a third six-inch site in Zurich planned for the first half of 2027.", "source_url": "https://futurumgroup.com/insights/coherent-q4-fy-2026-earnings-1-6t-transceivers-ramp-cpo-revenue-approaches/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/coherent", "permalink": "https://computefinder.net/entity/cf-sup-0053" }, { "id": "cf-sup-0054", "slug": "lumentum", "name": "Lumentum", "layer": "systems", "sublayer": "optics", "kind": "company", "hq_country": "US", "url": "https://www.lumentum.com/en", "summary": "Photonics company making lasers, optical transceivers and a 300 by 300 optical circuit switch for AI data centres.", "role": "A key laser supplier inside many optical modules, so its capacity affects transceiver availability.", "constraint": "TrendForce reports that Lumentum's chief executive said the indium phosphide supply and demand gap has passed that for DRAM and NAND.", "facts": [ { "text": "Lumentum's site lists a 200G PAM4 EML laser, an external laser source with ultra-high-power laser, a 1.6T 2xDR4 transceiver and the R300 optical circuit switch.", "source_url": "https://www.lumentum.com/en" }, { "text": "It is also named as a supplier in NVIDIA's March 2025 co-packaged optics switch announcement.", "source_url": "https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-Announces-Spectrum-X-Photonics-Co-Packaged-Optics-Networking-Switches-to-Scale-AI-Factories-to-Millions-of-GPUs/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/lumentum", "permalink": "https://computefinder.net/entity/cf-sup-0054" }, { "id": "cf-sup-0055", "slug": "innolight", "name": "InnoLight", "layer": "systems", "sublayer": "optics", "kind": "company", "hq_country": "CN", "url": "https://www.innolight.com/", "summary": "Suzhou-based optical transceiver maker founded in 2008, showing 800G and 1.6T pluggable modules for data centre and AI networks.", "role": "A major transceiver supplier to cloud and AI networks, named by NVIDIA among its optical transceiver partners.", "constraint": null, "facts": [ { "text": "InnoLight says it was founded in April 2008 in Suzhou and lists 800G modules and a 1.6T OSFP-XD DR8+ pluggable transceiver.", "source_url": "https://www.innolight.com/" }, { "text": "NVIDIA's March 2025 announcement lists Innolight among optical transceiver partners for its photonics switches.", "source_url": "https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-Announces-Spectrum-X-Photonics-Co-Packaged-Optics-Networking-Switches-to-Scale-AI-Factories-to-Millions-of-GPUs/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/innolight", "permalink": "https://computefinder.net/entity/cf-sup-0055" }, { "id": "cf-sup-0056", "slug": "ayar-labs", "name": "Ayar Labs", "layer": "systems", "sublayer": "optics", "kind": "company", "hq_country": "US", "url": "https://ayarlabs.com/", "summary": "San Jose company making TeraPHY optical engines for chip packages and the SuperNova remote light source, to link accelerators over optics instead of copper.", "role": "An optical I/O specialist aimed at rack-scale systems where chips are connected across racks by light.", "constraint": null, "facts": [ { "text": "Its site describes the TeraPHY optical engine as built into chip packages, and SuperNova as an ELSFP-format light source used alongside it.", "source_url": "https://ayarlabs.com/" }, { "text": "The site names AMD, NVIDIA, MediaTek, Alchip, GUC and Wiwynn among its backers.", "source_url": "https://ayarlabs.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/ayar-labs", "permalink": "https://computefinder.net/entity/cf-sup-0056" }, { "id": "cf-sup-0057", "slug": "lightmatter", "name": "Lightmatter", "layer": "systems", "sublayer": "optics", "kind": "company", "hq_country": "US", "url": "https://lightmatter.co/press-release/lightmatter-unveils-passage-m1000-photonic-superchip-worlds-fastest-ai-interconnect/", "summary": "Mountain View photonics company behind the Passage M1000 and L200 photonic interconnect products for linking many GPUs.", "role": "Develops photonic interposers that move data between chips optically, aimed at very large GPU domains.", "constraint": null, "facts": [ { "text": "A 31 March 2025 company release described Passage M1000 as a 3D active photonic interposer with 114 Tbps of total optical bandwidth over 256 fibres.", "source_url": "https://lightmatter.co/press-release/lightmatter-unveils-passage-m1000-photonic-superchip-worlds-fastest-ai-interconnect/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/lightmatter", "permalink": "https://computefinder.net/entity/cf-sup-0057" }, { "id": "cf-sup-0058", "slug": "indium-phosphide-laser-supply", "name": "Indium phosphide laser supply for optical modules", "layer": "systems", "sublayer": "optics", "kind": "bottleneck", "hq_country": null, "url": "https://trendforce.com/news/2026/08/06/news-inp-shortage-emerges-as-ai-optical-interconnect-bottleneck/", "summary": "Lasers for 800G and 1.6T optical modules depend on indium phosphide substrates, and trade press reports that supply of them is tight.", "role": "If lasers are scarce, optical transceivers for GPU clusters can be delayed, which can hold up cluster build-outs even when GPUs are in stock.", "constraint": "TrendForce (trade press, 6 August 2026) reports that analysts expect the supply imbalance to last beyond 2027 and that China accounts for about 70 per cent of refined indium, with export controls tightened since 2025.", "facts": [ { "text": "It is reported that Lumentum's chief executive said the indium phosphide supply gap has passed that of DRAM and NAND.", "source_url": "https://trendforce.com/news/2026/08/06/news-inp-shortage-emerges-as-ai-optical-interconnect-bottleneck/" }, { "text": "An analyst summary of Coherent's results reports that laser capacity, not assembly or test, limits its output, and that it plans to double indium phosphide output.", "source_url": "https://futurumgroup.com/insights/coherent-q4-fy-2026-earnings-1-6t-transceivers-ramp-cpo-revenue-approaches/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/indium-phosphide-laser-supply", "permalink": "https://computefinder.net/entity/cf-sup-0058" }, { "id": "cf-sup-0059", "slug": "vertiv", "name": "Vertiv", "layer": "facility", "sublayer": "cooling", "kind": "company", "hq_country": "US", "url": "https://www.vertiv.com/en-emea/about/news-and-insights/", "summary": "Vertiv supplies the power and thermal equipment that sits inside data halls, including liquid cooling, room and rack cooling, UPS, energy storage and power distribution.", "role": "It is a main source of the liquid cooling and critical power kit that dense AI halls need, so its product availability shapes how quickly new GPU capacity can be switched on.", "constraint": "Heat rejection equipment is a dependency for liquid-cooled halls, and Vertiv announced completing its purchase of heat-exchange maker ThermoKey in June 2026 to bring that capability in-house (company news page).", "facts": [ { "text": "Vertiv lists liquid cooling, thermal management, critical power (UPS, storage, DC power) and racks as its product areas for data centres.", "source_url": "https://www.vertiv.com/en-emea/about/news-and-insights/" }, { "text": "Vertiv reported completing the acquisition of ThermoKey S.p.A., a heat rejection and heat-exchange technology provider, on 15 June 2026.", "source_url": "https://www.vertiv.com/en-emea/about/news-and-insights/" }, { "text": "Vertiv describes itself as having roughly 34,000 employees and operations in more than 130 countries.", "source_url": "https://www.vertiv.com/en-us/about/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/vertiv", "permalink": "https://computefinder.net/entity/cf-sup-0059" }, { "id": "cf-sup-0060", "slug": "schneider-electric-data-centres", "name": "Schneider Electric (data centre solutions)", "layer": "facility", "sublayer": "cooling", "kind": "company", "hq_country": "FR", "url": "https://www.se.com/ww/en/work/solutions/data-centers-and-networks/", "summary": "Schneider Electric offers UPS systems, cooling (including liquid cooling), racks, modular pod and micro data centres, and management software for data centres.", "role": "It covers both the electrical and thermal sides of a facility, so it is relevant when judging whether a site can host high-density racks.", "constraint": null, "facts": [ { "text": "Its Galaxy UPS range is described as covering three-phase protection from 10 kW to 1,500 kW.", "source_url": "https://www.se.com/ww/en/work/solutions/data-centers-and-networks/" }, { "text": "Schneider lists liquid cooling as a cooling architecture for high-compute workloads, alongside its Uniflair cooling systems.", "source_url": "https://www.se.com/ww/en/work/solutions/data-centers-and-networks/" }, { "text": "The company also lists pod, modular, rack, micro and row data centre solutions and DCIM software.", "source_url": "https://www.se.com/ww/en/work/solutions/data-centers-and-networks/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/schneider-electric-data-centres", "permalink": "https://computefinder.net/entity/cf-sup-0060" }, { "id": "cf-sup-0061", "slug": "asetek", "name": "Asetek", "layer": "facility", "sublayer": "cooling", "kind": "company", "hq_country": "DK", "url": "https://www.asetek.com/", "summary": "Danish maker of liquid coolers that began with all-in-one coolers for PCs and has expanded into data centre cooling.", "role": "It is one of the direct-to-chip liquid cooling suppliers that server makers can draw on when building liquid-cooled GPU systems.", "constraint": null, "facts": [ { "text": "Asetek describes itself as a Danish company founded in 2000 and listed since 2013.", "source_url": "https://www.asetek.com/" }, { "text": "It says it invented the all-in-one liquid cooler and has expanded into data centre cooling solutions.", "source_url": "https://www.asetek.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/asetek", "permalink": "https://computefinder.net/entity/cf-sup-0061" }, { "id": "cf-sup-0062", "slug": "submer", "name": "Submer", "layer": "facility", "sublayer": "cooling", "kind": "company", "hq_country": null, "url": "https://submer.com/", "summary": "Submer provides immersion cooling and builds modular and AI-ready data centre facilities, and also offers GPU compute services through a related unit.", "role": "It spans cooling technology, prefabricated pods and facility development, which makes it a route to liquid-cooled capacity outside traditional colocation.", "constraint": null, "facts": [ { "text": "Submer states it has more than 500 MW of liquid-cooled infrastructure deployed and over 8 GW of deployment-ready capacity (company-stated figures).", "source_url": "https://submer.com/" }, { "text": "Its listed offerings include immersion cooling, the Rubix data centre service, SmartPod modular systems (EVO and EXO) and GPU services through Radian Arc.", "source_url": "https://submer.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/submer", "permalink": "https://computefinder.net/entity/cf-sup-0062" }, { "id": "cf-sup-0063", "slug": "green-revolution-cooling", "name": "Green Revolution Cooling (GRC)", "layer": "facility", "sublayer": "cooling", "kind": "company", "hq_country": "US", "url": "https://www.grcooling.com/", "summary": "Austin-based supplier of single-phase liquid immersion cooling tanks, fluids and management software for data centres and HPC.", "role": "It is an immersion cooling option for operators running dense AI or HPC hardware without traditional air cooling.", "constraint": null, "facts": [ { "text": "GRC says it has operated since 2009 with deployments in more than 20 countries.", "source_url": "https://www.grcooling.com/" }, { "text": "Its product line includes ICEraQ immersion systems, ICETank, a system manager and ElectroSafe fluids.", "source_url": "https://www.grcooling.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/green-revolution-cooling", "permalink": "https://computefinder.net/entity/cf-sup-0063" }, { "id": "cf-sup-0064", "slug": "eaton-switchgear-expansion", "name": "Eaton (switchgear manufacturing expansion)", "layer": "power", "sublayer": "electrical", "kind": "company", "hq_country": "US", "url": "https://www.businesswire.com/news/home/20260408952863/en/Eaton-Expands-Operations-in-Nebraska-with-New-Manufacturing-Facility-to-Meet-Increasing-Switchgear-Demand-Driven-by-AI-Data-Center-Boom", "summary": "Eaton supplies electrical equipment such as switchgear and power distribution, and announced a new medium-voltage switchgear plant to serve data centres, utilities and industry.", "role": "Medium-voltage switchgear connects a campus to the grid, so added manufacturing capacity bears on how fast powered sites can be energised.", "constraint": "The new Nebraska plant was announced in April 2026 with production reported to begin in the first half of 2027, so the added capacity is not yet available (company release).", "facts": [ { "text": "Eaton announced a 370,000 square foot medium-voltage switchgear facility in Bellevue, Nebraska, for air-insulated and gas-insulated products.", "source_url": "https://www.businesswire.com/news/home/20260408952863/en/Eaton-Expands-Operations-in-Nebraska-with-New-Manufacturing-Facility-to-Meet-Increasing-Switchgear-Demand-Driven-by-AI-Data-Center-Boom" }, { "text": "Production is stated to start in the first half of 2027, with hiring beginning later in 2026.", "source_url": "https://www.businesswire.com/news/home/20260408952863/en/Eaton-Expands-Operations-in-Nebraska-with-New-Manufacturing-Facility-to-Meet-Increasing-Switchgear-Demand-Driven-by-AI-Data-Center-Boom" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/eaton-switchgear-expansion", "permalink": "https://computefinder.net/entity/cf-sup-0064" }, { "id": "cf-sup-0065", "slug": "hitachi-energy-transformers", "name": "Hitachi Energy (transformers)", "layer": "power", "sublayer": "electrical", "kind": "company", "hq_country": "CH", "url": "https://www.hitachienergy.com/products-and-solutions/transformers", "summary": "Swiss-headquartered grid equipment maker covering power, distribution and special-application transformers, reactors and components.", "role": "Large transformers are needed to connect big data centre campuses to the transmission grid, and a small number of makers supply them.", "constraint": "Hitachi Energy said it was investing in expanded manufacturing capacity in China to help meet growing demand for grid equipment, which implies current capacity is a limit (company page).", "facts": [ { "text": "Hitachi Energy lists liquid-filled and dry-type, power, distribution, traction and special-application transformers in its portfolio.", "source_url": "https://www.hitachienergy.com/products-and-solutions/transformers" }, { "text": "It states it has supplied more than 20 UHVDC units at 800 kV and more than 500 AC units at 735 to 765 kV.", "source_url": "https://www.hitachienergy.com/products-and-solutions/transformers" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/hitachi-energy-transformers", "permalink": "https://computefinder.net/entity/cf-sup-0065" }, { "id": "cf-sup-0066", "slug": "us-transformer-lead-times", "name": "US transformer supply and lead times", "layer": "power", "sublayer": "electrical", "kind": "bottleneck", "hq_country": null, "url": "https://www.gao.gov/assets/d23106180.pdf", "summary": "A 2023 US Government Accountability Office report documented longer delivery times for large power and distribution transformers and heavy reliance on imports.", "role": "Long transformer waits can delay grid connection for new data centres, so they affect when announced capacity can actually come online.", "constraint": "The GAO reported lead times lengthening to years for some utilities, and said over 80 percent of large power transformers used in the US in 2019 were imported (GAO-23-106180, August 2023).", "facts": [ { "text": "One large utility told the GAO its transformer lead times had moved from 12 to 18 months to 18 to 36 months.", "source_url": "https://www.gao.gov/assets/d23106180.pdf" }, { "text": "The report cites a 2020 Commerce Department finding that more than 80 percent of large power transformers used in the US in 2019 were imported.", "source_url": "https://www.gao.gov/assets/d23106180.pdf" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/us-transformer-lead-times", "permalink": "https://computefinder.net/entity/cf-sup-0066" }, { "id": "cf-sup-0067", "slug": "ge-vernova-gas-turbines", "name": "GE Vernova (gas turbines)", "layer": "power", "sublayer": "on_site_power", "kind": "company", "hq_country": "US", "url": "https://www.gevernova.com/gas-power/products/gas-turbines", "summary": "GE Vernova makes heavy-duty and aeroderivative gas turbines, and markets aeroderivative units for powering data centres.", "role": "Gas turbines are a route to on-site or nearby power where grid connection is slow, so turbine supply affects where large AI campuses can be built.", "constraint": "Gas turbine delivery capacity is a reported constraint, with company-reported backlogs described in trade press (see the gas turbine backlog entry).", "facts": [ { "text": "GE Vernova lists heavy-duty models (such as 9HA, 7HA and 6F) and aeroderivative models (such as LM6000, LM2500 and TM2500).", "source_url": "https://www.gevernova.com/gas-power/products/gas-turbines" }, { "text": "Its page states the range spans 34 MW to 571 MW and says aeroderivative turbines can serve hyperscale and colocation data centres.", "source_url": "https://www.gevernova.com/gas-power/products/gas-turbines" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/ge-vernova-gas-turbines", "permalink": "https://computefinder.net/entity/cf-sup-0067" }, { "id": "cf-sup-0068", "slug": "siemens-energy-gas-turbines", "name": "Siemens Energy (gas turbines)", "layer": "power", "sublayer": "on_site_power", "kind": "company", "hq_country": "DE", "url": "https://www.siemens-energy.com/global/en/home/products-services/product-offerings/gas-turbines.html", "summary": "Siemens Energy supplies heavy-duty, industrial and aeroderivative gas turbines from small units up to large combined-cycle machines.", "role": "It is one of the few suppliers of large turbines and combined heat and power plants that data centre developers can turn to for dedicated power.", "constraint": "Reported order backlog at Siemens Energy and its peers is a delivery dependency for new gas-fired data centre power (see the gas turbine backlog entry).", "facts": [ { "text": "Siemens Energy states its gas turbine range runs from 2 MW to 593 MW, with the SGT5-9000HL at the top end.", "source_url": "https://www.siemens-energy.com/global/en/home/products-services/product-offerings/gas-turbines.html" }, { "text": "It says more than 2,500 aeroderivative units and over 2,250 industrial units are installed or sold worldwide.", "source_url": "https://www.siemens-energy.com/global/en/home/products-services/product-offerings/gas-turbines.html" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/siemens-energy-gas-turbines", "permalink": "https://computefinder.net/entity/cf-sup-0068" }, { "id": "cf-sup-0069", "slug": "mitsubishi-power-gas-turbines", "name": "Mitsubishi Power (gas turbine combined cycle)", "layer": "power", "sublayer": "on_site_power", "kind": "company", "hq_country": "JP", "url": "https://www.mhi.com/products/energy/gas_turbine_combined_cycle.html", "summary": "Mitsubishi Heavy Industries, through Mitsubishi Power, builds gas turbine combined cycle plants, including its J-series turbines.", "role": "It is a third large-turbine supplier alongside GE Vernova and Siemens Energy for projects needing dedicated gas-fired power.", "constraint": null, "facts": [ { "text": "Mitsubishi states its J-series combined cycle plants reach power generation efficiency of 64 percent or more (company claim).", "source_url": "https://www.mhi.com/products/energy/gas_turbine_combined_cycle.html" }, { "text": "The page does not make a data centre statement, though the site lists data centres as a separate product area.", "source_url": "https://www.mhi.com/products/energy/gas_turbine_combined_cycle.html" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/mitsubishi-power-gas-turbines", "permalink": "https://computefinder.net/entity/cf-sup-0069" }, { "id": "cf-sup-0070", "slug": "gas-turbine-backlog", "name": "Gas turbine backlogs", "layer": "power", "sublayer": "on_site_power", "kind": "bottleneck", "hq_country": null, "url": "https://www.turbomachinerymag.com/view/siemens-energy-posts-record-backlog-joining-ge-vernova-and-baker-hughes-in-a-record-quarter-for-gas-turbines", "summary": "Trade press reported in August 2026 that Siemens Energy, GE Vernova and Baker Hughes each posted record gas turbine related backlogs, with data centre demand named as a driver.", "role": "Long turbine order books mean gas-fired on-site power for new campuses may not be available quickly, which affects site and timing choices.", "constraint": "Figures are company-reported and relayed by Turbomachinery Magazine (6 August 2026), so they should be checked against the companies' own results.", "facts": [ { "text": "Siemens Energy was reported to have a gas services backlog of 73 billion euros, with US data centre orders cited as a driver.", "source_url": "https://www.turbomachinerymag.com/view/siemens-energy-posts-record-backlog-joining-ge-vernova-and-baker-hughes-in-a-record-quarter-for-gas-turbines" }, { "text": "GE Vernova was reported to have a gas turbine backlog of 116 GW.", "source_url": "https://www.turbomachinerymag.com/view/siemens-energy-posts-record-backlog-joining-ge-vernova-and-baker-hughes-in-a-record-quarter-for-gas-turbines" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/gas-turbine-backlog", "permalink": "https://computefinder.net/entity/cf-sup-0070" }, { "id": "cf-sup-0071", "slug": "bloom-energy", "name": "Bloom Energy", "layer": "power", "sublayer": "on_site_power", "kind": "company", "hq_country": "US", "url": "https://www.bloomenergy.com/news/bloom-energy-announces-gigawatt-fuel-cell-procurement-agreement-with-aep-to-power-ai-data-centers/", "summary": "San Jose company making solid oxide fuel cells that run on natural gas or hydrogen and can be installed at the customer site.", "role": "On-site fuel cells are one way to power a data centre while a grid connection is pending or limited.", "constraint": "The AEP agreement announced in November 2024 was an agreement of up to 1 GW with an initial 100 MW order, so most of that capacity is announced rather than delivered (company release).", "facts": [ { "text": "Bloom announced on 14 November 2024 a supply agreement with AEP of up to 1 GW of fuel cells for AI data centres, with an initial 100 MW order.", "source_url": "https://www.bloomenergy.com/news/bloom-energy-announces-gigawatt-fuel-cell-procurement-agreement-with-aep-to-power-ai-data-centers/" }, { "text": "The same release says Bloom had previously deployed more than 1.3 GW of fuel cells.", "source_url": "https://www.bloomenergy.com/news/bloom-energy-announces-gigawatt-fuel-cell-procurement-agreement-with-aep-to-power-ai-data-centers/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/bloom-energy", "permalink": "https://computefinder.net/entity/cf-sup-0071" }, { "id": "cf-sup-0072", "slug": "wartsila-data-centre-engines", "name": "Wärtsilä (data centre power engines)", "layer": "power", "sublayer": "on_site_power", "kind": "company", "hq_country": "FI", "url": "https://www.wartsila.com/media/news/20-11-2025-wartsila-continues-growth-in-the-data-center-segment-with-a-507-mw-order-in-the-us-offering-engines-as-a-reliable-power-solution-3686573", "summary": "Finnish engine and energy maker that supplies gas-fired reciprocating engine power plants, including for data centres.", "role": "Engine plants are a modular route to prime power on site, useful where grid supply is slow to arrive.", "constraint": "The 507 MW US order announced in November 2025 is stated for delivery in 2027, so it is an order, not operating capacity (company news).", "facts": [ { "text": "Wärtsilä announced a 507 MW order for 27 Wärtsilä 50SG engines for a US data centre under construction.", "source_url": "https://www.wartsila.com/media/news/20-11-2025-wartsila-continues-growth-in-the-data-center-segment-with-a-507-mw-order-in-the-us-offering-engines-as-a-reliable-power-solution-3686573" }, { "text": "The engines are to run on natural gas, with deliveries stated for 2027 and later conversion to sustainable fuels possible.", "source_url": "https://www.wartsila.com/media/news/20-11-2025-wartsila-continues-growth-in-the-data-center-segment-with-a-507-mw-order-in-the-us-offering-engines-as-a-reliable-power-solution-3686573" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/wartsila-data-centre-engines", "permalink": "https://computefinder.net/entity/cf-sup-0072" }, { "id": "cf-sup-0073", "slug": "caterpillar-hunt-energy", "name": "Caterpillar and Hunt Energy data centre power agreement", "layer": "power", "sublayer": "on_site_power", "kind": "programme", "hq_country": "US", "url": "https://dallasinnovates.com/caterpillar-and-hunt-energy-ink-agreement-to-deliver-1gw-of-data-center-power-across-north-america/", "summary": "A long-term agreement announced in August 2025 between Caterpillar and Hunt Energy to supply independent on-site power for data centres in North America.", "role": "It shows generator sets, turbines and batteries being packaged as dedicated power for data centres outside the grid queue.", "constraint": "The agreement is described as up to 1 GW with a first project in Texas, so it is reported as planned rather than operating (Dallas Innovates).", "facts": [ { "text": "Caterpillar is to supply gas and diesel generators, gas turbines, switchgear and controls, with Hunt adding battery storage.", "source_url": "https://dallasinnovates.com/caterpillar-and-hunt-energy-ink-agreement-to-deliver-1gw-of-data-center-power-across-north-america/" }, { "text": "The plan is up to 1 GW of generation capacity across North America, starting with a Texas project.", "source_url": "https://dallasinnovates.com/caterpillar-and-hunt-energy-ink-agreement-to-deliver-1gw-of-data-center-power-across-north-america/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/caterpillar-hunt-energy", "permalink": "https://computefinder.net/entity/cf-sup-0073" }, { "id": "cf-sup-0074", "slug": "fluence-data-centre-storage", "name": "Fluence (battery storage for data centres)", "layer": "power", "sublayer": "storage", "kind": "company", "hq_country": "US", "url": "https://info.fluenceenergy.com/hubfs/Fluence-Data%20Centre%20Brochure-BR-065-01-EN.pdf", "summary": "Battery storage supplier that markets its systems to data centres for backup, load smoothing and operating through grid connection delays.", "role": "Storage can absorb the sharp load swings of AI training and bridge delays in grid connection, which matters for siting new compute.", "constraint": null, "facts": [ { "text": "Fluence states it has more than 52 GWh of storage and over 300 projects globally as of March 2026 (company brochure).", "source_url": "https://info.fluenceenergy.com/hubfs/Fluence-Data%20Centre%20Brochure-BR-065-01-EN.pdf" }, { "text": "Stated data centre uses include replacing diesel backup, smoothing AI load variability and island operation during interconnection delays.", "source_url": "https://info.fluenceenergy.com/hubfs/Fluence-Data%20Centre%20Brochure-BR-065-01-EN.pdf" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/fluence-data-centre-storage", "permalink": "https://computefinder.net/entity/cf-sup-0074" }, { "id": "cf-sup-0075", "slug": "kairos-power-google-tva", "name": "Kairos Power, Google and TVA collaboration", "layer": "power", "sublayer": "nuclear", "kind": "programme", "hq_country": "US", "url": "https://world-nuclear-news.org/articles/google-kairos-power-tva-announce-collaboration", "summary": "A reported agreement under which TVA would buy power from Kairos Power's Hermes 2 reactor in Oak Ridge, Tennessee, to supply Google data centres in Tennessee and Alabama.", "role": "It is an early example of a utility, an advanced reactor developer and a hyperscaler tying new nuclear supply to data centre demand.", "constraint": "Operations are scheduled for 2030, so this is future supply and not currently operating (World Nuclear News report).", "facts": [ { "text": "Reported capacity is up to 50 MW delivered to the TVA grid, with construction permits issued in November 2024.", "source_url": "https://world-nuclear-news.org/articles/google-kairos-power-tva-announce-collaboration" }, { "text": "Kairos states its Hermes demonstration reactors are under construction in Oak Ridge with NRC permits secured.", "source_url": "https://kairospower.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/kairos-power-google-tva", "permalink": "https://computefinder.net/entity/cf-sup-0075" }, { "id": "cf-sup-0076", "slug": "oklo-aurora-inl", "name": "Oklo Aurora powerhouse at Idaho National Laboratory", "layer": "power", "sublayer": "nuclear", "kind": "programme", "hq_country": "US", "url": "https://www.power-eng.com/nuclear/oklo-taps-siemens-energy-for-steam-cycle-equipment-on-first-aurora-nuclear-project", "summary": "Oklo's planned 75 MWe sodium-cooled fast reactor at Idaho National Laboratory, the first of its Aurora powerhouses.", "role": "Oklo pitches long-term power agreements to customers including data centre operators, but this first plant is still pre-licence.", "constraint": "Reported target of initial operation in 2028 depends on licensing and other approvals, so no output should be assumed (Power Engineering).", "facts": [ { "text": "Oklo was reported to have broken ground at the Aurora INL site in September 2025, targeting initial operation in 2028 subject to licensing.", "source_url": "https://www.power-eng.com/nuclear/oklo-taps-siemens-energy-for-steam-cycle-equipment-on-first-aurora-nuclear-project" }, { "text": "Siemens Energy was reported to provide the steam-side power conversion system, and the NRC was reported to have accepted a design criteria topical report for review.", "source_url": "https://www.power-eng.com/nuclear/oklo-taps-siemens-energy-for-steam-cycle-equipment-on-first-aurora-nuclear-project" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/oklo-aurora-inl", "permalink": "https://computefinder.net/entity/cf-sup-0076" }, { "id": "cf-sup-0077", "slug": "terrapower-natrium", "name": "TerraPower Natrium (Kemmerer, Wyoming)", "layer": "power", "sublayer": "nuclear", "kind": "programme", "hq_country": "US", "url": "https://www.energy.gov/node/4856749", "summary": "A 345 MWe sodium-cooled fast reactor with molten salt storage, under construction in Kemmerer, Wyoming.", "role": "It is a regulator-cleared advanced reactor with storage that could in time supply large loads, though it is not itself a data centre project.", "constraint": "A separate NRC operating licence is still required before the plant can run, and completion is targeted for 2030 (US DOE).", "facts": [ { "text": "The NRC issued a construction permit on 9 March 2026, described by DOE as the first for a commercial non-light-water power reactor.", "source_url": "https://www.energy.gov/node/4856749" }, { "text": "The plant is 345 MWe and can boost output to 500 MW using molten salt storage.", "source_url": "https://www.energy.gov/node/4856749" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/terrapower-natrium", "permalink": "https://computefinder.net/entity/cf-sup-0077" }, { "id": "cf-sup-0078", "slug": "nuscale-power", "name": "NuScale Power", "layer": "power", "sublayer": "nuclear", "kind": "company", "hq_country": "US", "url": "https://www.nrc.gov/reactors/new-reactors/advanced/who-were-working-with/applicant-projects/nuscale-us460.html", "summary": "Developer of a light water small modular reactor whose uprated US460 design completed NRC standard design approval review.", "role": "It offers a licensed SMR design, with commercialisation through ENTRA1 and a collaboration with TVA, but no plant is reported operating for data centres.", "constraint": "A design approval is not a plant licence, so each site still needs its own licensing and construction before any power is delivered (NRC page).", "facts": [ { "text": "The NRC completed review of the US460 standard design approval in May 2025, about two months ahead of schedule.", "source_url": "https://www.nrc.gov/reactors/new-reactors/advanced/who-were-working-with/applicant-projects/nuscale-us460.html" }, { "text": "The design is six modules of 77 MWe each, 462 MWe in total.", "source_url": "https://www.nrc.gov/reactors/new-reactors/advanced/who-were-working-with/applicant-projects/nuscale-us460.html" }, { "text": "NuScale says ENTRA1 Energy is its exclusive commercialisation partner and that it is working with TVA on a 6 GW SMR programme.", "source_url": "https://www.nuscalepower.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/nuscale-power", "permalink": "https://computefinder.net/entity/cf-sup-0078" }, { "id": "cf-sup-0079", "slug": "x-energy-cascade-amazon", "name": "X-energy and Amazon Cascade facility", "layer": "power", "sublayer": "nuclear", "kind": "programme", "hq_country": "US", "url": "https://world-nuclear-news.org/articles/joint-venture-selected-to-deliver-cascade-smr-facility", "summary": "A planned small modular reactor plant near Richland, Washington, using X-energy Xe-100 reactors, developed by Energy Northwest with Amazon as funder and power purchaser.", "role": "It is a reported pairing of an operator-utility, a reactor vendor and a hyperscaler, still at planning and design stage.", "constraint": "Energy Northwest was reported to expect to submit an NRC construction permit application after the feasibility phase, so licensing is not yet complete (World Nuclear News).", "facts": [ { "text": "The initial phase is reported as four of twelve planned Xe-100 units, with operations targeted for the 2030s.", "source_url": "https://world-nuclear-news.org/articles/joint-venture-selected-to-deliver-cascade-smr-facility" }, { "text": "X-energy describes the Xe-100 as a commercial-scale advanced small modular reactor.", "source_url": "https://x-energy.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/x-energy-cascade-amazon", "permalink": "https://computefinder.net/entity/cf-sup-0079" }, { "id": "cf-sup-0080", "slug": "constellation-crane-restart", "name": "Constellation Crane Clean Energy Center restart", "layer": "power", "sublayer": "nuclear", "kind": "programme", "hq_country": "US", "url": "https://www.businesswire.com/news/home/20250624741471/en", "summary": "Constellation's plan to restart Unit 1 at the former Three Mile Island site, with Microsoft contracted to take the carbon-free power.", "role": "It shows an existing reactor being returned to service for a data centre customer, a different route from new-build SMRs.", "constraint": "The restart is described as possible as early as 2027 and relies on regulatory approvals and PJM interconnection, so it is not yet operating (Constellation release, June 2025).", "facts": [ { "text": "Constellation said it plans to restart Unit 1 as early as 2027, and that Microsoft will receive the plant's carbon-free electricity.", "source_url": "https://www.businesswire.com/news/home/20250624741471/en" }, { "text": "The release says PJM approved an expedited interconnection request for the project.", "source_url": "https://www.businesswire.com/news/home/20250624741471/en" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/constellation-crane-restart", "permalink": "https://computefinder.net/entity/cf-sup-0080" }, { "id": "cf-sup-0081", "slug": "stockholm-data-parks", "name": "Stockholm Data Parks", "layer": "facility", "sublayer": "heat_reuse", "kind": "programme", "hq_country": "SE", "url": "https://stockholmdataparks.com/", "summary": "A Stockholm initiative that links data centres to the city's district heating network so that waste heat is sold and reused.", "role": "It gives a concrete Nordic model for siting data centres where recovered heat has a buyer, which can matter for planning approval.", "constraint": "Heat reuse depends on an existing nearby district heating network, which Stockholm has but many regions lack (Stockholm Exergi via DCD).", "facts": [ { "text": "The initiative is led by the City of Stockholm, Stockholm Exergi, Ellevio and Stokab.", "source_url": "https://stockholmdataparks.com/" }, { "text": "A DigiPlex data centre was reported to connect to Stockholm Exergi's network to warm about 10,000 apartments, with nearly 90 percent of city buildings on district heating.", "source_url": "https://www.datacenterdynamics.com/en/news/digiplex-plugs-into-district-heating-in-stockholm/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/stockholm-data-parks", "permalink": "https://computefinder.net/entity/cf-sup-0081" }, { "id": "cf-sup-0082", "slug": "iea-energy-and-ai", "name": "IEA Energy and AI analysis", "layer": "power", "sublayer": "electrical", "kind": "programme", "hq_country": "FR", "url": "https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai", "summary": "International Energy Agency analysis estimating data centre electricity use now and in its base case to 2030.", "role": "It is a neutral reference for how much power data centres draw and why grid integration is a limiting factor.", "constraint": "The IEA notes data centres cluster in specific locations, which can make grid integration harder, and that energy infrastructure takes longer to build than data centres (IEA).", "facts": [ { "text": "The IEA estimates data centres used around 415 TWh in 2024, about 1.5 percent of global electricity.", "source_url": "https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai" }, { "text": "Its base case reaches around 945 TWh by 2030, just under 3 percent of global electricity.", "source_url": "https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/iea-energy-and-ai", "permalink": "https://computefinder.net/entity/cf-sup-0082" }, { "id": "cf-sup-0083", "slug": "ireland-cru-data-centre-connection-policy", "name": "Ireland CRU data centre connection policy", "layer": "power", "sublayer": "electrical", "kind": "programme", "hq_country": "IE", "url": "https://www.cru.ie/about-us/news/the-cru-publishes-its-decision-on-new-electricity-connection-policy-for-data-centres/", "summary": "Ireland's energy regulator decided in December 2025 on conditions for new large data centre grid connections.", "role": "It shows how a regulator can tie grid access to on-site or local generation and storage, which shapes where capacity can be built in Ireland.", "constraint": "Grid access depends on meeting the CRU's requirements, including generation or storage matching maximum import demand (CRU decision).", "facts": [ { "text": "The CRU decision of 12 December 2025 requires data centres to provide generation and/or storage, onsite or local, to match maximum import demand.", "source_url": "https://www.cru.ie/about-us/news/the-cru-publishes-its-decision-on-new-electricity-connection-policy-for-data-centres/" }, { "text": "It requires at least 80 percent of annual electricity from new Irish renewable projects, with a six-year development timeline allowed.", "source_url": "https://www.cru.ie/about-us/news/the-cru-publishes-its-decision-on-new-electricity-connection-policy-for-data-centres/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/ireland-cru-data-centre-connection-policy", "permalink": "https://computefinder.net/entity/cf-sup-0083" }, { "id": "cf-sup-0084", "slug": "ofgem-data-centre-connection-reform", "name": "Ofgem data centre connection reforms (GB)", "layer": "power", "sublayer": "electrical", "kind": "bottleneck", "hq_country": "GB", "url": "https://www.ofgem.gov.uk/consultation/proposed-data-centre-connection-reforms", "summary": "A consultation by the GB regulator on commitment fees and queue milestones for data centre grid connection applications.", "role": "Queue reform affects how quickly viable data centre projects in Great Britain can secure grid capacity.", "constraint": "Ofgem says non-viable projects in the connections queue are the problem being addressed; the measures are proposals, and the consultation closed on 17 September 2026 (Ofgem).", "facts": [ { "text": "Ofgem published the proposals on 29 July 2026, and the consultation closed on 17 September 2026.", "source_url": "https://www.ofgem.gov.uk/consultation/proposed-data-centre-connection-reforms" }, { "text": "The proposals include a non-refundable commitment fee and milestones covering a credible end user, long-lead electrical equipment and financial and technical capability.", "source_url": "https://www.ofgem.gov.uk/consultation/proposed-data-centre-connection-reforms" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/ofgem-data-centre-connection-reform", "permalink": "https://computefinder.net/entity/cf-sup-0084" }, { "id": "cf-sup-0085", "slug": "equinix", "name": "Equinix", "layer": "facility", "sublayer": "operators", "kind": "company", "hq_country": "US", "url": "https://www.equinix.com/company", "summary": "Global colocation and interconnection operator.", "role": "Retail colocation and cross-connects across many metros make it a point of entry for placing compute close to networks and cloud on-ramps.", "constraint": null, "facts": [ { "text": "Equinix states it operates 282 data centres across metros on six continents.", "source_url": "https://www.equinix.com/company" }, { "text": "It states more than 10,500 customers.", "source_url": "https://www.equinix.com/company" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/equinix", "permalink": "https://computefinder.net/entity/cf-sup-0085" }, { "id": "cf-sup-0086", "slug": "digital-realty", "name": "Digital Realty", "layer": "facility", "sublayer": "operators", "kind": "company", "hq_country": "US", "url": "https://www.digitalrealty.com/about", "summary": "Austin-headquartered data centre platform offering colocation and interconnection.", "role": "A widely distributed operator for hosting enterprise and AI deployments across many metros.", "constraint": null, "facts": [ { "text": "Digital Realty states more than 300 data centres in over 30 countries and over 55 metro areas.", "source_url": "https://www.digitalrealty.com/about" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/digital-realty", "permalink": "https://computefinder.net/entity/cf-sup-0086" }, { "id": "cf-sup-0087", "slug": "iron-mountain-data-centers", "name": "Iron Mountain Data Centers", "layer": "facility", "sublayer": "operators", "kind": "company", "hq_country": "US", "url": "https://www.ironmountain.com/data-centers", "summary": "Colocation arm of Iron Mountain with sites across North America, Europe and Asia-Pacific, including several Indian cities.", "role": "An option for hyperscale and retail colocation, including in India and Singapore.", "constraint": null, "facts": [ { "text": "Iron Mountain states a global 1.4 GW portfolio across three continents and more than 30 locations.", "source_url": "https://www.ironmountain.com/data-centers" }, { "text": "Listed sites include Amsterdam, Frankfurt, London, Madrid, Singapore and Indian cities such as Mumbai, Chennai and Hyderabad.", "source_url": "https://www.ironmountain.com/data-centers" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/iron-mountain-data-centers", "permalink": "https://computefinder.net/entity/cf-sup-0087" }, { "id": "cf-sup-0088", "slug": "airtrunk", "name": "AirTrunk", "layer": "facility", "sublayer": "operators", "kind": "company", "hq_country": null, "url": "https://www.airtrunk.com/", "summary": "Hyperscale data centre platform in Asia-Pacific and the Middle East.", "role": "A wholesale option for large deployments in markets such as Australia, Japan, India, Malaysia and Saudi Arabia.", "constraint": null, "facts": [ { "text": "AirTrunk lists campuses in Australia, Hong Kong, India, Japan, Malaysia, Singapore and Saudi Arabia.", "source_url": "https://www.airtrunk.com/" }, { "text": "It mentions expanding in Johor Bahru, Malaysia, to 700 MW or more through new facilities (company statement).", "source_url": "https://www.airtrunk.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/airtrunk", "permalink": "https://computefinder.net/entity/cf-sup-0088" }, { "id": "cf-sup-0089", "slug": "stack-infrastructure", "name": "STACK Infrastructure", "layer": "facility", "sublayer": "operators", "kind": "company", "hq_country": null, "url": "https://www.stackinfra.com/", "summary": "Developer and operator of hyperscale data centres in the Americas, Asia-Pacific and EMEA.", "role": "A wholesale option across 24 markets for cloud and AI capacity.", "constraint": null, "facts": [ { "text": "STACK lists 24 markets across the Americas, Asia-Pacific and EMEA.", "source_url": "https://www.stackinfra.com/" }, { "text": "It states more than 8.5 GW built and under development and more than 7.5 GW planned (company figures).", "source_url": "https://www.stackinfra.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/stack-infrastructure", "permalink": "https://computefinder.net/entity/cf-sup-0089" }, { "id": "cf-sup-0090", "slug": "vantage-data-centers", "name": "Vantage Data Centers", "layer": "facility", "sublayer": "operators", "kind": "company", "hq_country": "US", "url": "https://www.vantage-dc.com/", "summary": "Wholesale data centre campus provider in North America, EMEA and Asia-Pacific.", "role": "Supplies large powered campuses for hyperscalers and AI tenants.", "constraint": null, "facts": [ { "text": "Vantage states campuses range from 64 MW to over 1 GW.", "source_url": "https://www.vantage-dc.com/" }, { "text": "It lists a consortium of backers including DigitalBridge, Silver Lake, AustralianSuper and PSP Investments.", "source_url": "https://www.vantage-dc.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/vantage-data-centers", "permalink": "https://computefinder.net/entity/cf-sup-0090" }, { "id": "cf-sup-0091", "slug": "africa-data-centres", "name": "Africa Data Centres", "layer": "facility", "sublayer": "operators", "kind": "company", "hq_country": null, "url": "https://www.africadatacentres.com/", "summary": "Pan-African colocation and cloud-connectivity operator, part of Cassava Technologies.", "role": "One of the few operators with facilities in several African countries, relevant to local or regional capacity.", "constraint": null, "facts": [ { "text": "It lists facilities in Johannesburg, Cape Town, Nairobi and Lagos.", "source_url": "https://www.africadatacentres.com/" }, { "text": "It names Cassava Technologies as its parent group.", "source_url": "https://www.africadatacentres.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/africa-data-centres", "permalink": "https://computefinder.net/entity/cf-sup-0091" }, { "id": "cf-sup-0092", "slug": "q-ant", "name": "Q.ANT", "layer": "alternative", "sublayer": "photonic", "kind": "company", "hq_country": "DE", "url": "https://qant.com", "summary": "A Stuttgart company building photonic processors that do certain mathematical operations with light rather than transistors. Its Native Processing Server is sold as a photonic compute product, backed by an open-source software kit.", "role": "A route to lower-energy AI and simulation maths, but today a specialist accelerator reached mainly through research centres.", "constraint": "Hardware access appears limited: the company page says software written with its kit will run on the photonic hardware as access opens.", "maturity": "early_commercial", "facts": [ { "text": "The company's page says its photonic servers are deployed at two German research institutions, LRZ and Jülich.", "source_url": "https://qant.com" }, { "text": "An open-source software development kit with C/C++ and Python interfaces and a simulation backend is listed as publicly available, with a launch noted for September 2026.", "source_url": "https://qant.com" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/q-ant", "permalink": "https://computefinder.net/entity/cf-sup-0092" }, { "id": "cf-sup-0093", "slug": "spinncloud", "name": "SpiNNcloud", "layer": "alternative", "sublayer": "neuromorphic", "kind": "company", "hq_country": "DE", "url": "https://www.spinncloud.com", "summary": "A German company offering brain-inspired computing systems built on the SpiNNaker2 chip, aimed at AI inference that exploits sparse activity.", "role": "A neuromorphic option for inference workloads where sparsity can be exploited, sold as systems rather than rented by the hour.", "constraint": "The page names no customers, so its statement that institutions in Europe and the US use the systems cannot be checked from it.", "maturity": "early_commercial", "facts": [ { "text": "The company describes SpiNNaker2 as commercially available and lists a successor, SpiNNext, as coming soon.", "source_url": "https://www.spinncloud.com" }, { "text": "It states that its systems are already used by institutions in Europe and the US (self-reported).", "source_url": "https://www.spinncloud.com" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/spinncloud", "permalink": "https://computefinder.net/entity/cf-sup-0093" }, { "id": "cf-sup-0094", "slug": "intel-hala-point", "name": "Intel Hala Point", "layer": "alternative", "sublayer": "neuromorphic", "kind": "technology", "hq_country": "US", "url": "https://www.businesswire.com/news/home/20240417915986/en", "summary": "A large neuromorphic research system built from Intel Loihi 2 chips and first installed at Sandia National Laboratories in the United States.", "role": "Shows how far spiking, event-driven hardware has been scaled, which matters as a pointer to future low-power inference, not as capacity to book today.", "constraint": "Intel itself calls it a research prototype that will inform future commercial systems (Intel announcement, April 2024).", "maturity": "research", "facts": [ { "text": "The system combines 1,152 Loihi 2 processors, giving about 1.15 billion artificial neurons across 140,544 cores.", "source_url": "https://www.businesswire.com/news/home/20240417915986/en" }, { "text": "It was first deployed at Sandia National Laboratories for research on brain-scale computing.", "source_url": "https://www.businesswire.com/news/home/20240417915986/en" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/intel-hala-point", "permalink": "https://computefinder.net/entity/cf-sup-0094" }, { "id": "cf-sup-0095", "slug": "brainchip-akida", "name": "BrainChip Akida", "layer": "alternative", "sublayer": "neuromorphic", "kind": "company", "hq_country": "US", "url": "https://www.brainchip.com", "summary": "A neuromorphic processor family for small, always-on AI that runs on the device itself. It is offered as chips, as licensable processor cores and with developer kits.", "role": "A purchasable low-power option for edge sensing workloads where data should stay on the device.", "constraint": "The use cases listed on the company page are keyword spotting, vision classification and sensor anomaly detection, not large-model serving.", "maturity": "commercial", "facts": [ { "text": "The company lists its AKD1500 chip, in PCIe and M.2 forms, as shipping.", "source_url": "https://www.brainchip.com" }, { "text": "Akida processor cores can be licensed for custom silicon, and a software kit, model zoo and developer hardware are available through a developer hub.", "source_url": "https://www.brainchip.com" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/brainchip-akida", "permalink": "https://computefinder.net/entity/cf-sup-0095" }, { "id": "cf-sup-0096", "slug": "ibm-analog-ai", "name": "IBM Research analog AI", "layer": "alternative", "sublayer": "in_memory", "kind": "technology", "hq_country": "US", "url": "https://research.ibm.com/projects/analog-ai", "summary": "An IBM research programme on analogue in-memory computing, where neural-network weights are held in memory devices that also do the arithmetic, cutting data movement.", "role": "A long-range route to lower-energy inference; it signals where accelerator design may go rather than offering capacity now.", "constraint": "The project page describes prototype chips and research results only; it lists no product or customer access.", "maturity": "research", "facts": [ { "text": "IBM's inference work uses phase-change memory cells, and the page says its chips hold over 13 million of them.", "source_url": "https://research.ibm.com/projects/analog-ai" }, { "text": "The page records prototype chips in 2023 that used much less power for AI tasks and software-equivalent accuracy on transformer networks in 2021.", "source_url": "https://research.ibm.com/projects/analog-ai" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/ibm-analog-ai", "permalink": "https://computefinder.net/entity/cf-sup-0096" }, { "id": "cf-sup-0097", "slug": "cerebras", "name": "Cerebras", "layer": "alternative", "sublayer": "in_memory", "kind": "company", "hq_country": "US", "url": "https://www.cerebras.ai", "summary": "Builds AI processors that use an entire silicon wafer as one chip, with large on-chip memory, and sells access to them as a cloud service, a dedicated private service or on-premises systems.", "role": "A non-GPU architecture available now for fast model inference, useful where GPU supply or latency is a problem.", "constraint": "Classed here under in-memory only because of its large on-chip memory; the register has no separate wafer-scale category.", "maturity": "commercial", "facts": [ { "text": "The company page lists three ways to use its systems: public cloud endpoints with an API key, dedicated private cloud, and on-premises deployment.", "source_url": "https://www.cerebras.ai" }, { "text": "Its chip page describes the current wafer-scale engine as 46,225 square millimetres with 900,000 cores and a design that tolerates manufacturing defects by routing around them.", "source_url": "https://www.cerebras.ai/chip" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/cerebras", "permalink": "https://computefinder.net/entity/cf-sup-0097" }, { "id": "cf-sup-0098", "slug": "ibm-quantum", "name": "IBM Quantum", "layer": "alternative", "sublayer": "quantum", "kind": "programme", "hq_country": "US", "url": "https://www.ibm.com/quantum/access-plans", "summary": "IBM's cloud service for running programs on its quantum computers, with a free tier, pay-as-you-go and reserved access, plus an on-premises option.", "role": "One of the few quantum systems that anyone can reach today through a normal account, for experiments and research rather than general computing.", "constraint": "The free Open plan has a monthly runtime allowance, so larger work needs a paid plan (IBM access page).", "maturity": "commercial", "facts": [ { "text": "IBM lists five access routes: a free Open plan, pay-as-you-go, Flex, Premium and On-Prem.", "source_url": "https://www.ibm.com/quantum/access-plans" }, { "text": "A credits programme offers extra compute time to eligible researchers, and classroom accounts let teachers manage student access.", "source_url": "https://www.ibm.com/quantum/access-plans" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/ibm-quantum", "permalink": "https://computefinder.net/entity/cf-sup-0098" }, { "id": "cf-sup-0099", "slug": "ionq", "name": "IonQ", "layer": "alternative", "sublayer": "quantum", "kind": "company", "hq_country": "US", "url": "https://ionq.com", "summary": "A trapped-ion quantum computing company whose machines can be reached through the major cloud platforms or its own cloud.", "role": "A second route to real quantum hardware without owning any, using accounts people may already hold with a cloud provider.", "constraint": null, "maturity": "commercial", "facts": [ { "text": "The company says its systems can be used through Amazon Braket, Microsoft Azure Quantum and Google Cloud, or directly through IonQ Quantum Cloud.", "source_url": "https://ionq.com" }, { "text": "Its site names several system generations, including Tempo, Forte, Forte Enterprise and Aria.", "source_url": "https://ionq.com" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/ionq", "permalink": "https://computefinder.net/entity/cf-sup-0099" }, { "id": "cf-sup-0100", "slug": "nvidia-jetson", "name": "NVIDIA Jetson", "layer": "alternative", "sublayer": "edge", "kind": "technology", "hq_country": "US", "url": "https://developer.nvidia.com/embedded-computing", "summary": "A line of small, power-efficient computer modules and developer kits for running AI in robots, cameras and other devices away from a data centre.", "role": "A common building block for on-site and on-device inference where sending data to the cloud is impractical.", "constraint": null, "maturity": "commercial", "facts": [ { "text": "NVIDIA describes Jetson as hardware modules plus a software stack for edge AI and robotics, sold as developer kits and production modules.", "source_url": "https://developer.nvidia.com/embedded-computing" }, { "text": "The page lists Jetson Thor and the Orin Nano Super developer kit among current products.", "source_url": "https://developer.nvidia.com/embedded-computing" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/nvidia-jetson", "permalink": "https://computefinder.net/entity/cf-sup-0100" }, { "id": "cf-sup-0101", "slug": "akash-network", "name": "Akash Network", "layer": "alternative", "sublayer": "decentralised", "kind": "programme", "hq_country": null, "url": "https://akash.network", "summary": "An open marketplace where independent providers bid to host containerised workloads, including GPU jobs, in a reverse auction run by the buyer's requirements.", "role": "A way to rent capacity from independent hosts instead of a single cloud, with the buyer choosing between competing bids.", "constraint": "The network page itself shows a modest pool, 62 providers and 423 GPUs when fetched (self-reported), so large reservations may not fit.", "maturity": "early_commercial", "facts": [ { "text": "Users can run models through an API service, deploy containers via a console where providers bid, or join as a provider.", "source_url": "https://akash.network" }, { "text": "The site reported 62 active providers, 423 GPUs and 15,000 vCPUs on 2 October 2026 (self-reported).", "source_url": "https://akash.network" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/akash-network", "permalink": "https://computefinder.net/entity/cf-sup-0101" }, { "id": "cf-sup-0102", "slug": "io-net", "name": "io.net", "layer": "alternative", "sublayer": "decentralised", "kind": "company", "hq_country": null, "url": "https://io.net", "summary": "A platform that aggregates GPUs from many suppliers and offers them as on-demand clusters, with models and containers deployable on top.", "role": "A pooled GPU supply for AI teams that want cluster-style access from distributed hardware.", "constraint": "Its capacity and country-count claims are self-reported and the page gives no headquarters or dates.", "maturity": "early_commercial", "facts": [ { "text": "The site says it offers instant access to more than 30,000 GPUs and clusters in over 130 countries (self-reported).", "source_url": "https://io.net" }, { "text": "Deployment options named are containers, Ray clusters and bare metal.", "source_url": "https://io.net" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/io-net", "permalink": "https://computefinder.net/entity/cf-sup-0102" }, { "id": "cf-sup-0103", "slug": "render-network", "name": "Render Network", "layer": "alternative", "sublayer": "decentralised", "kind": "programme", "hq_country": null, "url": "https://rendernetwork.com", "summary": "A distributed GPU network where node operators contribute spare capacity for 3D rendering and creative AI work, run by OTOY with a separate nonprofit foundation for the protocol.", "role": "Pay-as-you-go GPU capacity aimed at rendering and image or video generation rather than general training.", "constraint": "The supported tools on the page are rendering engines and creative AI services, so it is not a general-purpose GPU rental.", "maturity": "commercial", "facts": [ { "text": "The page says OTOY operates the platform while the Render Network Foundation, a nonprofit, advances the protocol and community.", "source_url": "https://rendernetwork.com" }, { "text": "Supported render engines include OctaneRender, Redshift and Blender Cycles, with on-demand billing and no minimum commitment.", "source_url": "https://rendernetwork.com" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/render-network", "permalink": "https://computefinder.net/entity/cf-sup-0103" }, { "id": "cf-sup-0104", "slug": "boinc", "name": "BOINC and Science United", "layer": "alternative", "sublayer": "decentralised", "kind": "technology", "hq_country": "US", "url": "https://boinc.berkeley.edu", "summary": "Open-source software that lets people donate spare computer time to scientific projects, based at the University of California, Berkeley, with Science United as a simple way to pick research areas.", "role": "The long-running model of volunteer compute; useful for researchers who can package work as many small jobs, not for buying capacity.", "constraint": "Capacity comes from volunteers for science projects, so it cannot be rented on demand (BOINC site).", "maturity": "commercial", "facts": [ { "text": "The site says BOINC and Science United are located at UC Berkeley and supported by the US National Science Foundation.", "source_url": "https://boinc.berkeley.edu" }, { "text": "Roughly 30 science projects are stated to use BOINC.", "source_url": "https://boinc.berkeley.edu" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/boinc", "permalink": "https://computefinder.net/entity/cf-sup-0104" }, { "id": "cf-sup-0105", "slug": "google-project-suncatcher", "name": "Google Project Suncatcher", "layer": "alternative", "sublayer": "space", "kind": "programme", "hq_country": "US", "url": "https://research.google/blog/exploring-a-space-based-scalable-ai-infrastructure-system-design/", "summary": "A Google research study into constellations of solar-powered satellites carrying AI accelerators and linked by optical connections. Announced as research, not a service.", "role": "A proposal for compute in orbit that is not available to anyone; relevant only as a signal of where large operators are looking for power and space.", "constraint": "Google says the economics depend on launch costs falling substantially by the mid-2030s, and a prototype mission is only planned (Google research blog).", "maturity": "proposed", "facts": [ { "text": "Google describes the work as research with a published preprint and plans a prototype mission with Planet Labs of two satellites by early 2027.", "source_url": "https://research.google/blog/exploring-a-space-based-scalable-ai-infrastructure-system-design/" }, { "text": "The blog reports a bench test of inter-satellite optical links at 800 Gbps each way and radiation testing of its Trillium TPUs.", "source_url": "https://research.google/blog/exploring-a-space-based-scalable-ai-infrastructure-system-design/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/google-project-suncatcher", "permalink": "https://computefinder.net/entity/cf-sup-0105" }, { "id": "cf-sup-0106", "slug": "starcloud", "name": "Starcloud", "layer": "alternative", "sublayer": "space", "kind": "company", "hq_country": "US", "url": "https://blogs.nvidia.com/blog/starcloud/", "summary": "A company that put a data-centre GPU on a small satellite as a test of orbital computing. Reported by trade press as launched in November 2025; larger orbital data centres are proposals only.", "role": "An early demonstration of orbital compute; there is no public service for buyers.", "constraint": "Trade press reports an expected mission life of about 11 months at low orbit, and the larger multi-gigawatt facility is only a proposal (DatacenterDynamics).", "maturity": "prototype", "facts": [ { "text": "NVIDIA's blog describes Starcloud-1 as a roughly 60 kg satellite carrying an H100 GPU, intended to run an open language model and process Earth-observation data in orbit.", "source_url": "https://blogs.nvidia.com/blog/starcloud/" }, { "text": "DatacenterDynamics reported that the satellite launched on 3 November 2025 and that the company's chief executive said it was operating normally (reported).", "source_url": "https://www.datacenterdynamics.com/en/news/starcloud-1-satellite-reaches-space-with-nvidia-h100-gpu-now-operating-in-orbit/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/starcloud", "permalink": "https://computefinder.net/entity/cf-sup-0106" }, { "id": "cf-sup-0107", "slug": "2africa", "name": "2Africa subsea cable", "layer": "connectivity", "sublayer": "subsea", "kind": "programme", "hq_country": null, "url": "https://engineering.fb.com/2025/11/17/connectivity/core-2africa-system-completion-future-connectivity/", "summary": "A Meta-led submarine cable system around Africa linking Africa, Europe and, with its Pearls segment, parts of the Middle East and South Asia.", "role": "Adds international capacity to landing stations across Africa, which bears on where data centres and edge sites can be reached.", "constraint": "Meta's November 2025 post says the Pearls segment was still scheduled to go live in 2026, so its status should be rechecked.", "maturity": "commercial", "facts": [ { "text": "Meta's November 2025 post says the core system is complete and describes the whole system as about 45,000 km.", "source_url": "https://engineering.fb.com/2025/11/17/connectivity/core-2africa-system-completion-future-connectivity/" }, { "text": "The Pearls segment adds landings in the Gulf states, Pakistan, India and others; Meta's 2021 announcement said the full system would connect 33 countries.", "source_url": "https://engineering.fb.com/2021/09/28/connectivity/2africa-pearls/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/2africa", "permalink": "https://computefinder.net/entity/cf-sup-0107" }, { "id": "cf-sup-0108", "slug": "equiano", "name": "Equiano subsea cable", "layer": "connectivity", "sublayer": "subsea", "kind": "programme", "hq_country": "US", "url": "https://blog.google/intl/en-africa/company-news/inside-google/the-equiano-subsea-cable-has-arrived-in/", "summary": "A Google cable running from Portugal down the Atlantic coast of Africa to South Africa, with branches serving Togo, Nigeria, Namibia and St Helena.", "role": "A high-capacity western route to Africa, relevant to data centres in Lagos, Windhoek and the Cape Town area.", "constraint": "Trade press describes onward capacity being sold through local landing-station partners rather than by Google directly (DatacenterDynamics, Paratus report).", "maturity": "commercial", "facts": [ { "text": "Google's post names Portugal, Togo, Nigeria, Namibia, South Africa and St Helena on the route and records the cable arriving in Togo in March 2022.", "source_url": "https://blog.google/intl/en-africa/company-news/inside-google/the-equiano-subsea-cable-has-arrived-in/" }, { "text": "Trade press reports a design capacity of 144 Tbit/s over 12 fibre pairs, and that the Namibian branch was activated on 12 June 2023.", "source_url": "https://www.datacenterdynamics.com/en/news/paratus-group-activates-namibian-branch-of-equiano-cable/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/equiano", "permalink": "https://computefinder.net/entity/cf-sup-0108" }, { "id": "cf-sup-0109", "slug": "project-waterworth", "name": "Meta Project Waterworth", "layer": "connectivity", "sublayer": "subsea", "kind": "programme", "hq_country": "US", "url": "https://engineering.fb.com/2025/02/14/connectivity/project-waterworth-ai-subsea-infrastructure/", "summary": "A cable project announced by Meta in February 2025 to connect the US, India, Brazil, South Africa and other regions over a route of more than 50,000 km. Announced, not in service.", "role": "A proposed new set of ocean routes with unusually many fibre pairs, framed by Meta around supporting AI infrastructure.", "constraint": "Meta describes it as a multi-year build under development and gives no service date.", "maturity": "proposed", "facts": [ { "text": "Meta says the system will be over 50,000 km long and reach five continents.", "source_url": "https://engineering.fb.com/2025/02/14/connectivity/project-waterworth-ai-subsea-infrastructure/" }, { "text": "The design uses 24 fibre pairs, compared with 8 to 16 in comparable systems according to Meta.", "source_url": "https://engineering.fb.com/2025/02/14/connectivity/project-waterworth-ai-subsea-infrastructure/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/project-waterworth", "permalink": "https://computefinder.net/entity/cf-sup-0109" }, { "id": "cf-sup-0110", "slug": "telegeography-submarine-cable-map", "name": "TeleGeography Submarine Cable Map", "layer": "connectivity", "sublayer": "subsea", "kind": "technology", "hq_country": "US", "url": "https://www.submarinecablemap.com", "summary": "A free interactive map of submarine cable systems and their landing stations, published by the telecoms research firm TeleGeography.", "role": "The quickest way to see which cables land near a prospective data-centre site.", "constraint": "The site describes itself as covering major systems, so smaller cables may be absent.", "maturity": "commercial", "facts": [ { "text": "The site describes itself as a regularly updated interactive map of the world's major submarine cable systems and landing stations.", "source_url": "https://www.submarinecablemap.com/about" }, { "text": "TeleGeography is named as its publisher.", "source_url": "https://www.submarinecablemap.com" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/telegeography-submarine-cable-map", "permalink": "https://computefinder.net/entity/cf-sup-0110" }, { "id": "cf-sup-0111", "slug": "submarine-cable-resilience", "name": "Submarine cable damage and slow restoration", "layer": "connectivity", "sublayer": "subsea", "kind": "bottleneck", "hq_country": null, "url": "https://www.itu.int/digital-resilience/submarine-cables/advisory-body/terms-of-reference/", "summary": "Nearly all intercontinental traffic runs on undersea cables, and the ITU and the International Cable Protection Committee set up an advisory body to tackle slow licensing, installation and repair and the risk of physical damage.", "role": "A reminder that data-centre connectivity to remote regions depends on a small number of physical cables that can take time to restore.", "constraint": "The ITU terms of reference cite licensing, installation and restoration timelines and physical damage as the problems to address.", "maturity": "commercial", "facts": [ { "text": "The ITU page says submarine cables carry more than 99 per cent of intercontinental connectivity.", "source_url": "https://www.itu.int/digital-resilience/submarine-cables/advisory-body/terms-of-reference/" }, { "text": "The advisory body, set up by the ITU and ICPC, aims to speed licensing, installation and restoration and to protect cables from damage.", "source_url": "https://www.itu.int/digital-resilience/submarine-cables/advisory-body/terms-of-reference/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/submarine-cable-resilience", "permalink": "https://computefinder.net/entity/cf-sup-0111" }, { "id": "cf-sup-0112", "slug": "starlink", "name": "Starlink", "layer": "connectivity", "sublayer": "satellite", "kind": "company", "hq_country": "US", "url": "https://www.starlink.com/business", "summary": "A low-Earth-orbit satellite internet service from SpaceX that offers connections at home, on the move and for businesses.", "role": "A way to connect remote or temporary edge sites that have no fibre, at latency suited to many but not all workloads.", "constraint": null, "maturity": "commercial", "facts": [ { "text": "Starlink's own page says it provides fast internet at home and on the go, including in rural and remote places. The site is script-driven and little else could be read from it.", "source_url": "https://www.starlink.com/business" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/starlink", "permalink": "https://computefinder.net/entity/cf-sup-0112" }, { "id": "cf-sup-0113", "slug": "amazon-leo", "name": "Amazon Leo", "layer": "connectivity", "sublayer": "satellite", "kind": "company", "hq_country": "US", "url": "https://www.aboutamazon.com/news/amazon-leo", "summary": "Amazon's low-Earth-orbit satellite network, formerly Project Kuiper, being built out and offered first to selected business and carrier partners.", "role": "A second large LEO option for remote and edge sites as the constellation is built out.", "constraint": "Amazon's news page shows partnerships and launches but not a full public commercial launch, so availability varies by country and customer.", "maturity": "early_commercial", "facts": [ { "text": "Amazon's news page lists more than 375 satellites in orbit and partnerships including Herotel in South Africa and Uztelecom in Uzbekistan.", "source_url": "https://www.aboutamazon.com/news/amazon-leo" }, { "text": "IEEE ComSoc's tech blog reported an enterprise preview from late 2025 with production terminals shipped to selected companies (reported).", "source_url": "https://techblog.comsoc.org/2025/11/25/amazon-leo-formerly-project-kuiper-unveils-satellite-broadband-for-enterprises-competitive-analysis-with-starlink/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/amazon-leo", "permalink": "https://computefinder.net/entity/cf-sup-0113" }, { "id": "cf-sup-0114", "slug": "eutelsat-oneweb", "name": "Eutelsat OneWeb", "layer": "connectivity", "sublayer": "satellite", "kind": "company", "hq_country": "FR", "url": "https://www.eutelsat.com/", "summary": "A low-Earth-orbit constellation run by Eutelsat for enterprise, government, aviation and maritime links, sold alongside the operator's geostationary fleet.", "role": "A business-focused LEO network for connecting remote sites, including high-latitude ones.", "constraint": "The page points customers to a partner portal, which suggests service is bought through partners rather than directly.", "maturity": "commercial", "facts": [ { "text": "Eutelsat says its OneWeb LEO constellation has more than 600 satellites and offers low-latency links pole to pole.", "source_url": "https://www.eutelsat.com/" }, { "text": "The page lists video contribution, emergency communications, aviation, maritime and general enterprise connectivity as uses.", "source_url": "https://www.eutelsat.com/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/eutelsat-oneweb", "permalink": "https://computefinder.net/entity/cf-sup-0114" }, { "id": "cf-sup-0115", "slug": "mlperf-inference", "name": "MLCommons MLPerf Inference", "layer": "alternative", "sublayer": "open_hardware", "kind": "programme", "hq_country": "US", "url": "https://mlcommons.org/benchmarks/inference-datacenter/", "summary": "An industry benchmark suite, run by MLCommons, that measures how fast systems run trained models and also records system power in some scenarios.", "role": "A shared, published yardstick for comparing accelerators and systems on speed and energy use.", "constraint": "Entries are split into Available, Preview and research categories, so not every result is hardware a buyer can obtain (MLCommons page).", "maturity": "commercial", "facts": [ { "text": "The suite uses latency, throughput, server and offline scenarios, in a closed division for like-for-like comparison and an open division for innovation.", "source_url": "https://mlcommons.org/benchmarks/inference-datacenter/" }, { "text": "Power is measured at the wall for the whole system; the page, last modified on 21 September 2026, lists round v6.1.", "source_url": "https://mlcommons.org/benchmarks/inference-datacenter/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/mlperf-inference", "permalink": "https://computefinder.net/entity/cf-sup-0115" }, { "id": "cf-sup-0116", "slug": "uk-ai-research-resource", "name": "UK AI Research Resource (AIRR)", "layer": "alternative", "sublayer": "sovereign", "kind": "programme", "hq_country": "GB", "url": "https://www.gov.uk/government/publications/ai-research-resource/airr-advanced-supercomputers-for-the-uk", "summary": "A UK government programme giving researchers, academics and small businesses access to AI supercomputers at Bristol and Cambridge, with application routes that provide compute but no funding.", "role": "A public, application-based source of GPU time for UK projects.", "constraint": "Applications must show benefit to the UK, and awarded time must be used within three months of project start (GOV.UK and UKRI pages).", "maturity": "commercial", "facts": [ { "text": "GOV.UK lists Isambard-AI at Bristol with 5,448 NVIDIA GH200 chips, and Dawn and Zenith at Cambridge.", "source_url": "https://www.gov.uk/government/publications/ai-research-resource/airr-advanced-supercomputers-for-the-uk" }, { "text": "Rapid Access offers UK small and medium businesses up to 20,000 GPU hours, and the Gateway route offers first-time users up to 10,000 GPU hours.", "source_url": "https://www.gov.uk/government/publications/ai-research-resource/airr-advanced-supercomputers-for-the-uk" }, { "text": "UKRI shows the Gateway route open with no closing date as of its February 2026 update, with compute only and no funding.", "source_url": "https://www.ukri.org/opportunity/access-to-isambard-ai-and-dawn-airr-supercomputers-gateway-route/" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/uk-ai-research-resource", "permalink": "https://computefinder.net/entity/cf-sup-0116" }, { "id": "cf-sup-0117", "slug": "eu-ai-factories", "name": "EU AI Factories", "layer": "alternative", "sublayer": "sovereign", "kind": "programme", "hq_country": null, "url": "https://digital-strategy.ec.europa.eu/en/policies/ai-factories", "summary": "A European Commission initiative, run through the EuroHPC Joint Undertaking, that pairs AI-oriented supercomputers with data and support services in national consortia.", "role": "A route for European startups, researchers and public bodies to publicly backed AI compute and help using it.", "constraint": "The Commission page frames access as being for European users from industry, research, academia and public authorities.", "maturity": "early_commercial", "facts": [ { "text": "The Commission lists 19 AI Factories and 13 associated Antennas, selected in waves from December 2024 to October 2025.", "source_url": "https://digital-strategy.ec.europa.eu/en/policies/ai-factories" }, { "text": "EuroHPC oversees operation and access, and the page says startups and small businesses are prioritised.", "source_url": "https://digital-strategy.ec.europa.eu/en/policies/ai-factories" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/eu-ai-factories", "permalink": "https://computefinder.net/entity/cf-sup-0117" }, { "id": "cf-sup-0118", "slug": "indiaai-compute-portal", "name": "IndiaAI Compute Portal", "layer": "alternative", "sublayer": "sovereign", "kind": "programme", "hq_country": "IN", "url": "https://compute.indiaai.gov.in", "summary": "An Indian government portal giving eligible researchers, startups, students and public bodies access to GPU cloud services from empanelled providers at reduced cost.", "role": "A national pooled-GPU scheme that shows how a state can aggregate private data-centre capacity for public use.", "constraint": "Access depends on eligibility criteria such as startup recognition or a research record, and requests over 5,000 GPU hours need committee review (portal page).", "maturity": "early_commercial", "facts": [ { "text": "The portal page describes four steps: login, user verification, a compute request with a cost estimate, and approval, with requests under 5,000 GPU hours approved automatically.", "source_url": "https://compute.indiaai.gov.in" }, { "text": "Trade press citing the ministry reported more than 38,000 GPUs onboarded as of March 2026 (reported).", "source_url": "https://telecomtalk.info/?p=1005751" } ], "checked_on": "2026-10-02", "page_url": "https://computefinder.net/supply-chain/indiaai-compute-portal", "permalink": "https://computefinder.net/entity/cf-sup-0118" } ] }