How much Amazon invests
In the second quarter of 2026, Amazon spent $54.2 billion on equipment and buildings, the company said in its quarterly report. This is capital expenditure: money for data centers and equipment that serve for years.
For the full year 2026, Forbes estimated in October 2026 that Amazon’s capex would be about $200 billion. In the same estimate, Microsoft spends about $190 billion, Alphabet $180–190 billion and Meta $125–145 billion.
Together, the four companies will invest $700–725 billion in 2026, according to Forbes, against $410 billion in 2025, about +75%. Amazon is first on this list by amount.
The money returns through the AWS cloud, which sells computing to other companies. In the second quarter of 2026, AWS revenue grew 37% to $42.2 billion.
A CFO should read these numbers together: $54.2 billion of quarterly investment and $42.2 billion of quarterly cloud revenue. They cannot be compared one to one: equipment bought in a quarter works for years, while revenue comes every quarter.
On 30 July 2026, the company raised its 2026 capex forecast to $220 billion from $200 billion, Fortune reported. Amazon named more expensive memory as the reason, and Andy Jassy added that capacity will still not cover all of this year’s demand.
Read the October Forbes figure as the outlet’s estimate, and the July company forecast as Amazon’s own plan. The date when all numbers were checked is at the bottom of the page, next to the sources.
Where the money goes: infrastructure and models
Amazon invests in AWS cloud data centers and in its own Trainium AI chips. A major client for these chips is Anthropic, an AI model developer.
According to Anthropic, the company trains and runs its Claude models on more than 1 million Amazon Trainium2 chips.
Under an agreement of 20 April 2026, almost 1 GW of capacity on Trainium2 and Trainium3 chips should start working by the end of 2026. Over time, the agreement provides for up to 5 GW.
A gigawatt here is a measure of the electrical capacity of data centers: the market measures the size of AI infrastructure in gigawatts instead of the number of servers.
Amazon’s scheme works like this: the company builds capacity and makes chips, and the model developer commits to using them. According to Anthropic, it has committed to spend more than $100 billion on AWS technology over 10 years.
For Amazon, this is an order for years ahead; for Anthropic, secured access to computing. The risk of such a link is that the supplier’s revenue depends on the money the client raises in funding rounds.
For a company that buys an AI solution, the lesson is the same: the longer and larger the commitment to a computing supplier, the more important it is to set the price and the exit terms in advance.
Stakes in other companies
Amazon has two large AI stakes known from reports and the press: in Anthropic and in OpenAI.
The Anthropic stake already shows in the accounts. In the second quarter of 2026, Amazon received $53.4 billion of other income before tax, mostly from the revaluation of its Anthropic stake.
A revaluation is not cash in the bank: it is a change in the paper value of the stake after new Anthropic deals. A CFO reads this line of the report separately from revenue and operating profit.
On 20 April 2026, Amazon invested another $5 billion in Anthropic and may add up to $20 billion, Anthropic said.
The second stake appeared in April 2026. Amazon became a lead investor in OpenAI’s $122 billion round at a valuation of $852 billion, with $50 billion, according to Bloomberg Law.
About $35 billion of this sum depends on an IPO, OpenAI’s first public share offering, or on a technology goal. Nvidia and SoftBank joined Amazon as lead investors in the round, with $30 billion each.
So Amazon holds stakes in two model developers that compete with each other, and it does not depend on the success of one model.
Why giants buy stakes in AI companies and what the risks are is covered on the page about how big tech invests in startups.
Deals of 2026
Amazon’s two large deals of 2026, by date:
1. April 2026, OpenAI. A $122 billion round at a valuation of $852 billion: Amazon $50 billion, of which about $35 billion is conditional, Nvidia $30 billion, SoftBank $30 billion (Bloomberg Law).
2. 20 April 2026, Anthropic. Amazon invested $5 billion with the right to add up to $20 billion, and Anthropic committed to spend more than $100 billion on AWS over 10 years (Anthropic website).
Both deals are built in a similar way: most of the money is conditional. In OpenAI, about $35 billion of $50 billion depends on a stock market listing or a technology goal; in Anthropic, up to $20 billion may be added later.
This gives a practical rule for a news reader: a headline with a large sum shows the maximum. How much money moved at once is visible only in the deal terms.
The Anthropic deal also works both ways: Amazon invests in a stake, and Anthropic commits to spend money on the AWS cloud for 10 years and to get almost 1 GW of capacity by the end of 2026.
All the large deals of the year, with amounts, dates and sources, are collected in the deal tracker.
What a company can learn
The four largest companies will spend $700–725 billion on capital expenditure in 2026, by Forbes’ estimate. The sums are on another scale, but three moves from Amazon’s deals work at any budget:
1. Separate promised money from money paid. Of Amazon’s $50 billion in OpenAI, about $35 billion is conditional; in a contract for an AI solution, also pay in parts after results.
2. Keep investment and revenue on separate lines. Amazon invested $54.2 billion in a quarter, and AWS revenue grew 37%: judge the effect of AI by the money that comes in regularly, not by the purchase amount.
3. Fix the price and the right to exit. Anthropic committed to spend more than $100 billion on AWS over 10 years; in your contract with a contractor, fix the price of running the model and the right to move the solution to another supplier.
A mid-sized company does not need to build capacity: it needs to find the processes where ready-made AI returns money and to calculate the effect before the first spending.
You can start with the AI roadmap: it shows where the company loses money, which solutions close the gap and what they will bring in money.