Home / Wiki / H / Hyperscaler

Hyperscaler

Technology and VendorsUpdated 16 Nov 2026

A hyperscaler is a technology company that builds and operates data centres on a very large scale and delivers computing, storage and networking services from them to customers around the world. The term comes from hyperscale computing, an approach to system design in which capacity is expanded by adding large numbers of standard servers that are managed by software as a single pool.

The name is applied most often to the three largest providers of public cloud computing: Amazon Web Services, Microsoft Azure and Google Cloud. It is also used for companies that run comparable infrastructure mainly for their own services, such as Meta, and for other large cloud operators including Oracle and, in China, Alibaba and Tencent. There is no formal definition, but the companies concerned each operate many sites with hundreds of thousands of servers or more.

01

Business model

A hyperscaler rents out computing resources on demand. Customers pay for what they use instead of buying and running their own hardware, and can add or release capacity within minutes. Services range from raw processing and storage to databases, analytics and development tools, the last of which are described under platform as a service. Many software companies build their products on top of this infrastructure.

The model depends on economies of scale. Buying servers, chips, land and electricity in vast quantities lowers unit costs, and spreading the cost of software and engineering over a huge base lowers them further. The largest operators design their own servers, networking equipment and processors. The capital required, and the difficulty customers face in moving systems from one provider to another, are significant barriers to entry, and the market has the structure of an oligopoly.

02

Capital spending and artificial intelligence

Hyperscalers are among the largest corporate investors in the world. Their capital expenditure goes on land and buildings, power and cooling systems, servers, storage and networks. Since the arrival of generative AI, spending has risen sharply to build each AI data center needed to train and run large models. These facilities rely on a specialised graphics processing unit or custom accelerator chip in place of general-purpose processors and draw far more electricity per rack.

A simple measure of the commitment is capital intensity:

Capital intensity=Capital expenditureRevenue\text{Capital intensity} = \frac{\text{Capital expenditure}}{\text{Revenue}}

A company that spends 60 billion dollars in a year on revenue of 300 billion dollars has a capital intensity of 60÷300=2060 \div 300 = 20 percent. A rise in the ratio shows that investment is growing faster than sales, which puts pressure on free cash flow until the new capacity earns revenue.

Electricity has become a binding constraint. Operators sign long-term supply contracts, including each power purchase agreement with a wind, solar or nuclear generator, and compete for sites with grid connections.

03

Why hyperscalers matter to investors

  • Index weight. The parent companies are among the largest listed companies in the world, so their results weigh heavily on broad share indices and on technology funds.
  • Supplier demand. Their spending plans drive orders for semiconductor makers, server and networking vendors, electrical equipment firms, builders and utilities. Analysts track quarterly spending guidance as an indicator for the whole AI infrastructure supply chain.
  • Property and finance. Hyperscalers lease as well as own facilities, which makes them the main tenants of data centre landlords, including some organised as a real estate investment trust, and a growing presence in bond and private credit markets.
  • Cloud growth. The growth rate and operating margin of each cloud division are among the most closely watched figures in quarterly earnings.
04

Risks and criticism

The main financial risk is that capacity is built ahead of demand. Investment in computing has moved in cycles before, and an AI infrastructure cycle in which revenue from new services falls short of expectations would leave operators with underused assets. Those assets wear out or become obsolete quickly. The depreciation period chosen for servers and chips has a large effect on reported profit, and changes to it are examined closely by analysts.

Other concerns include the concentration of customers among a few AI developers, arrangements in which a hyperscaler invests in a company that then spends the money on its cloud, and the use of leases and joint ventures that keep some obligations off the balance sheet. Supporters of continued spending cite the Jevons paradox: as computing becomes cheaper, total demand for it tends to rise.

Regulators have examined the sector for competition problems, including fees for moving data out of a provider's cloud and software licensing terms that favour a provider's own platform. Dependence on a small number of providers is also treated as an operational risk for banks and other regulated firms, and the large electricity and water needs of data centres have drawn environmental objections and local opposition.

Last updated 16 Nov 2026