MMOOZZGG.com

AI investment

Big tech capex 2026: AI spending by the world’s largest companies

6 min read4 sources

Big tech capex in 2026 for Amazon, Microsoft, Alphabet and Meta will reach $700–725 billion, against $410 billion in 2025. That is about +75% in a year, by Forbes’ count. Below are the figures per company, what the money goes to and the risks analysts see.

On this page · 7 sections

$700 billion in a year

Capex, that is, money for buildings, servers, chips and networks that serve for years, will reach $700–725 billion for the four companies in 2026. In 2025 this sum was $410 billion.

By company: Amazon about $200 billion, Microsoft about $190 billion, Alphabet $180–190 billion, Meta $125–145 billion. Forbes gives these figures; for Alphabet and Meta they are ranges, not exact sums.

By an estimate quoted in an industry review, the capex of hyperscalers, that is, the companies with the largest clouds, grew by about 80% in a year. Next year more than $1 trillion is expected there.

The two growth estimates, about +75% from Forbes and about 80% in the industry review, count different sets of companies. But they reach the same conclusion: spending almost doubled in a year.

The expectation of more than $1 trillion next year remains an expectation, not a commitment. Companies revise their capex plans in quarterly reports, and we update the figures on this page from them.

For scale: Gartner forecasts that the whole world will spend $2.59 trillion on AI in 2026. The four companies make up a notable part of this sum with their capex.

Amazon

Amazon will spend about $200 billion on capex in 2026, more than any other company of the four.

Most of the money goes to the AWS cloud: data centers, servers and its own chips for training and running models. Amazon rents this capacity to companies that build AI services and earns on renting out computing.

For a cloud customer this means: there will be capacity for AI, but the rental price depends on how fast Amazon earns back its investment. A computing contract with a price fixed for a year protects your budget from such changes.

Amazon’s bet is simple: the more companies launch AI, the more cloud capacity they need. Details of the company’s AI investments, including its stakes in model developers, are on the Amazon page.

Microsoft

Microsoft will put about $190 billion into capex in 2026, almost as much as Amazon.

The money goes to the Azure cloud and to AI features in office and enterprise software. Microsoft sells AI mainly to large companies, so it builds capacity for their workloads and their data storage requirements.

The vendor earns back capex through customer subscriptions. So the price of AI features in familiar work software is directly linked to these investments.

A separate item: computing for model developers that Microsoft is tied to by contracts. How the company’s AI investments are built is covered on the Microsoft page.

Alphabet

Alphabet, the parent company of Google, plans capex of $180–190 billion in 2026. The company named a range, not an exact sum.

Alphabet has three areas that drive spending: the Google Cloud, its own AI chips and its own models, which are built into search and other services.

The three companies with clouds, Amazon, Microsoft and Alphabet, spend comparable sums: $180 to $200 billion each. The gap between them is smaller than the gap to Meta.

Its own chips let Alphabet depend less on purchases from outside manufacturers. The company’s AI investments, with sources, are on the Alphabet and Google page.

Meta

Meta plans $125–145 billion of capex in 2026. Its range is wider than that of the other three companies.

Unlike the other three, Meta does not rent cloud capacity to outside customers. The money goes to its own models and to recommendation and advertising systems in the company’s products.

Even the lower bound, $125 billion, puts Meta on a par with the cloud companies, although it does not plan to sell computing to outside customers.

Such spending without a cloud business shows that Meta sees its own models as a condition for its products, not as a separate line of business with revenue.

So Meta expects these investments to pay back not through renting out computing but through advertising revenue. The figures for the company are on the Meta page.

Where the money goes

Gartner estimates global spending on AI infrastructure in 2026 at $1.43 trillion out of $2.59 trillion of total AI spending. The four companies’ capex is part of this infrastructure.

1. Chips and servers. Most of the hardware for training models is bought from Nvidia and other chipmakers; the companies make some of it themselves.

2. Data centers. Buildings, cooling systems and grid connections are built for years ahead.

3. Networks. Models are trained on thousands of chips at once, and the links between them cost money of their own.

For 2027 Gartner forecasts $3.49 trillion of AI spending, against $1.76 trillion in 2025. How this money is split between hardware and software is covered on the page about AI infrastructure.

For a Russian company there is no direct link to these purchases, but there is an indirect one. Global demand for chips and servers affects the price and delivery times of the equipment you install on site.

Will the investments pay off?

There is no answer yet, and we do not make forecasts. There are risk estimates worth watching.

A Roscongress report put the probability of a crash of the AI company market in 2026 at more than 50%. According to the report, the S&P 500 index has grown by about $7.5 trillion since early 2024, and $4.9 trillion of this growth came from 17 AI-related companies.

The same report describes circular financing: chipmakers invest in customers who buy their hardware. In this scheme, part of the demand for capacity is created by the suppliers’ own money.

If demand for AI services grows more slowly than data centers are built, companies will wait longer for a return on what they built. Signs of an overheated market are covered on the page about the risks of AI investment.

Capex pays back over years. A data center building serves for decades, while servers and chips become outdated in a few years. If the return comes later than the equipment ages, part of the investment will have to be written off.

For a Russian company, the lesson is not the scale but the order of steps. The giants build capacity for expected demand; for smaller companies it pays to first find the processes where AI will surely return the money, and only then pay for hardware.

Sources and fact-checking

Checked by the editors on 14 October 2026.

  1. Forbes, Jason Kirsch, 7 October 2026 (forbes.com)
  2. Yandex Cloud blog, webinar “A look at the AI market”, 24 July 2026 (yandex.cloud)
  3. Gartner, press release, 19 May 2026 (gcom.pdo.aws.gartner.com)
  4. Mail Finance (finance.mail.ru), citing RIA Novosti and a Roscongress report, 19 February 2026 (finance.mail.ru)

Found a mistake? Tell us

Publications

Contact us