What is a hyperscaler?

A hyperscaler is a cloud provider that builds and operates computing infrastructure at global scale and rents it to anyone as self-service, on-demand services. The term usually means the three largest: Amazon Web Services (AWS), Microsoft Azure and Google Cloud, which together accounted for 61% of worldwide spending on cloud infrastructure services in the third quarter of 2026. Oracle Cloud Infrastructure and Alibaba Cloud are sometimes counted as hyperscalers too.

What makes a provider a hyperscaler

There is no formal threshold. The word describes scale along four dimensions at once:

  • Footprint. Dozens of regions worldwide, each split into several availability zones with separate power and networking.
  • Breadth. Hundreds of managed services (databases, queues, identity, analytics, AI) behind one account, one API and one bill.
  • Capital spending. Hyperscalers build their own data centers, design their own chips, such as AWS Graviton, Google Axion and TPUs, and Azure Cobalt, and buy power years ahead.
  • Self-service. Anyone with a payment card can create resources through an API in minutes, without talking to sales.

A provider with a large footprint but a narrow catalogue, or a broad catalogue in a handful of locations, is usually not called a hyperscaler.

Hyperscale data centers and hyperscale operators

“Hyperscale” also describes buildings. Synergy Research Group counted 1,360 large data centers operated by hyperscale companies at the end of 2025, holding 48% of the world’s data center capacity. Its tracking covers 21 cloud and internet companies, including Meta, Apple, ByteDance and Tencent.

Those companies run hyperscale data centers for their own services but do not rent general-purpose infrastructure to the public. “Hyperscale operator” is therefore a wider group than “hyperscaler” in the sense a cloud buyer means: a provider you can open an account with.

Who counts, and market share

Synergy’s Q3 2026 cloud market update, published on 30 October 2026, put spending on cloud infrastructure services at an estimated $167.7 billion for the quarter, up 55% year on year.

ProviderShare of Q3 2026 spendingOwn Arm server CPU
Amazon Web Services26%Graviton
Microsoft Azure19%Cobalt
Google Cloud16%Axion
Oracle Cloud InfrastructureNot given; named among the fastest-growing tier-two providersNone

Synergy also reported that neoclouds, GPU-focused providers such as CoreWeave, are growing fast enough to erode some established providers’ shares. Market shares cover infrastructure services (IaaS, PaaS and hosted private cloud), not SaaS revenue such as Microsoft 365.

AI spending

Most hyperscaler investment now goes into data centers, power and accelerators for AI. Amazon alone expects about $220 billion of cash capital expenditure in 2026, as stated in its third-quarter results on 29 October 2026. Spending at that scale is one reason GPU capacity is still sold out in places, as explained in why it is hard to get GPU capacity in the cloud.

Hyperscalers compared with other provider types

HyperscalerNeocloudRegional or European cloudBare-metal and VPS host
ExamplesAWS, Azure, Google CloudCoreWeave, Lambda, NebiusOVHcloud, Scaleway, IONOS, STACKITHetzner, Leaseweb
Main productFull cloud platformGPU capacity for AIVMs, storage, managed Kubernetes, some databasesPhysical or virtual servers
Managed servicesHundredsA few, AI-focusedTensFew or none
LocationsGlobal, multiple availability zones per regionSelected GPU sitesA few countries, usually in EuropeA few sites
Outbound trafficBilled per GBVariesOften included or cheaperUsually included up to a quota

A neocloud competes with hyperscalers on GPU price and availability, covered in what a neocloud is and how it differs from a hyperscaler. European providers compete on jurisdiction and price, covered in which European cloud providers can run Kubernetes.

What the hyperscaler premium buys

Hyperscaler list prices for compute and especially for outbound traffic are higher than those of smaller providers. A side-by-side example is in how Hetzner Cloud compares to AWS. In return you get:

  • Managed services you do not have to run. Databases, queues, identity and monitoring are operated for you.
  • Availability zones. Several isolated data centers per region let you survive a data center failure without leaving the region.
  • Global reach. Regions on every continent for latency and data residency.
  • Compliance paperwork. Long lists of audits and certifications that enterprise customers and regulators ask for.
  • Commercial programs. Startup credits, committed-spend discounts and marketplaces that let customers buy third-party software against their cloud commitment.

The same breadth is what makes leaving expensive. Applications written against proprietary services, per-GB egress fees and multi-year commitments all raise the cost of moving, as explained in what cloud egress fees are. For organizations in the EU, the three largest hyperscalers are US companies, which raises the question of whether the US CLOUD Act applies to data stored in the EU.

Using hyperscalers alongside other providers

Many teams use a hyperscaler for the managed services they depend on and run compute-heavy or traffic-heavy workloads on cheaper providers. Cloudfleet Kubernetes Engine (CFKE) runs one cluster whose node auto-provisioning creates nodes in your own AWS and Google Cloud accounts, on Hetzner Cloud and on your own servers. See how Fleets work.

On this page