Hyperscale cloud instances
Virtual machines and bare metal from very large general-purpose clouds, with accelerators beside many other services.
Definition, not a sourced statistic
What it is
A hyperscale cloud rents virtual machines, and sometimes whole physical servers, from a very large general-purpose platform. Accelerator instances are one product among storage, databases, networking, identity and many managed services, all billed through one account. You choose a region, start an instance and the provider runs everything underneath it.
Who it suits
- Teams whose data and applications already sit on that cloud, so compute is close to them.
- Organisations that need one contract, one bill and one security review across many services.
- Work that grows and shrinks, where self-serve start and stop matters more than a fixed cluster.
Common contract shapes
- On-demand. Start and stop when you like, billed for the time used, with no promise the accelerator type is free when you ask.
- Spot or preemptible. Spare capacity the provider can take back at short notice, on separate terms.
- Reserved, savings plan or committed use. A commitment over a term in return for a different rate. Check whether it is a billing discount or a physical set-aside of machines.
- Enterprise agreement. A negotiated contract covering spend across many services, often with its own support and data terms.
Providers use their own names for these shapes, so compare what is promised, not what it is called. The contract shapes guide explains each one.
What to ask
- Egress and data transfer. What is charged to move data out to the internet, to another region and to another provider, and is any of it waived?
- Service levels. Does the service level agreement cover the accelerator instance itself, or only the wider region, and what credit does a breach give?
- Data residency and access. Which regions offer the accelerator you need, and are backups, logs and support access kept in the same place?
- Security certifications. Which certifications cover the exact service and region you will use, and can you see the reports under a non-disclosure agreement?
- Export controls and sanctions. Are some accelerator types or regions closed to your country, sector or end use, and who checks that?
- Exit. How do you export machine images, data and configuration, and what happens to committed spend if you leave early?
- Is quota for the accelerator granted on request, and how long does an increase take?
The full list is on the printable buyer's checklist. For rules on chips and where they can go, see export controls and compute thresholds.
On this site
Entries listed here because their own summary or checked facts on this site fit this type. An entry can fit more than one type, and a provider missing here may still offer it. This is a list, not a ranking, and in name order. We link to providers for prices.
Providers
- Alibaba Cloud: A global cloud platform with elastic virtual servers including GPU instances, and the Model Studio AI service.
- Amazon Web Services (EC2): The virtual server service of a very large general-purpose cloud. It covers a wide catalogue of instance types, including accelerated options for machine learning.
- Google Cloud (Compute Engine): Google's virtual machine service. Besides GPUs, it is the place to reach Google's own Cloud TPU accelerators for machine learning.
- Huawei Cloud: A global cloud platform with virtual servers, GPU-accelerated servers, the ModelArts machine learning platform and the Pangu family of large models.
- IBM Cloud: IBM's public cloud, offering GPU and AI accelerator options including NVIDIA GPUs, Intel Gaudi 3 and AMD Instinct MI300X, with bare metal and OpenShift integration.
- Microsoft Azure: Microsoft's cloud virtual machine service, with GPU-enabled and HPC-oriented machine types alongside general-purpose ones.
- NAVER Cloud Platform: A South Korean cloud platform offering infrastructure, platform and software services for corporations, with GPU servers and the CLOVA AI services.
- Oracle Cloud Infrastructure: Oracle's cloud compute service, including bare metal servers and GPU instances with both NVIDIA and AMD accelerators.
Related
- Glossary: Hyperscaler, On-demand, Spot, Reserved, Egress, SLA (service level agreement), Data residency.
- Guide: How to get GPUs this week
- Guide: What is a capacity block, and the other contract shapes
- Provider directory
- Compare offer types
Other offer types: GPU clouds and bare metal · Marketplaces and aggregators · Decentralised and community networks · Regional and sovereign clouds · Managed inference and serverless GPU · Chip-maker and accelerator clouds · National and institutional HPC, and free or funded programmes · Colocation, wholesale, powered shell and build-to-suit · On-premises hardware, and leasing or financing it · Capacity brokers, reserved contracts, secondary and spot markets · Edge and telecom sites, and converted sites
Definitions and questions written in our own words, checked on 3 October 2026 and due for review by 3 January 2027. Sourced facts stay on the provider, programme and policy pages this page links to. This is general information to help you ask better questions. It is not legal, tax or procurement advice. For export controls, sanctions and contracts, check the primary rules and talk to a qualified adviser. Something wrong? Tell us.