Amazon expects to spend about $220 billion on capital projects in 2026, raising its forecast by $20 billion after a quarter in which Amazon Web Services posted its fastest growth in 18 quarters.
The company disclosed the new figure on July 30. AWS sales rose 37% in the April-June period, while Amazon’s overall net sales increased 20% to $200.6 billion. The spending plan includes more than AI data centers: Amazon is also investing in semiconductors, robotics and satellites. But the expansion of AI and cloud capacity is the central force behind the number.
At this scale, capital expenditure is no longer a support function for the product strategy. It is the strategy. The companies able to secure chips, power, land and financing can make more computing available, lower unit costs and attract the demand needed to justify the next build.
The physical layer sets the pace
Generative AI is often presented as software that improves at digital speed. Its limiting inputs are stubbornly physical. A data center can operate for decades, while the servers inside it turn over through several generations. Grid connections, transmission equipment and power generation move on still longer schedules.
That mismatch creates a forecasting problem. Amazon must commit capital before all of the future demand is visible. If it builds too slowly, customers wait for capacity and move workloads elsewhere. If it builds too quickly, expensive assets arrive before enough revenue is ready to use them.
The latest AWS growth gives Amazon evidence for the aggressive case. It also raises the bar: faster revenue growth has to continue long enough to support a balance sheet carrying unprecedented infrastructure commitments. Free cash flow can weaken while cash is converted into assets whose returns arrive over years.
A global race with different models
Amazon is not building in isolation. Microsoft, Alphabet, Meta and Nvidia are directing enormous sums toward computing, custom chips and data-center partnerships. Europe is taking a different route, using €10 billion in public funding to encourage seven AI gigafactories and at least €20 billion more in private investment.
The comparison is revealing. U.S. hyperscalers are building proprietary networks at a scale few companies can match. Europe is trying to create shared capacity as industrial policy. Both approaches recognize the same fact: access to compute now shapes who can develop models, which businesses can afford to use them and where the resulting economic value accumulates.
The infrastructure race also moves costs outside the cloud bill. Utilities and communities have to manage demand for electricity, water and transmission. Regions trade tax incentives and faster permitting for investment, jobs and a place in the AI supply chain. The political durability of the buildout will depend on whether local benefits keep pace with local burdens.
What enterprise buyers should watch
More capacity should eventually improve availability and price competition. It can also deepen dependence on a small group of platforms. Enterprises should evaluate portability, committed-spend terms, regional capacity and the ability to move inference among models and providers.
The strategic question is not whether Amazon’s number is too large in the abstract. It is what assumptions the number makes about the future of work, software and demand. A $220 billion plan says Amazon believes AI consumption will become a core utility of the economy – and that owning the utility layer will be one of its most valuable positions.
About Julie: Julie Wohlberg is a journalist, communications strategist and entrepreneur with more than 25 years of experience building brands and launching companies. A former Tribune news and feature writer and early executive at Fotolia, she writes about media, brand, consumer behavior and the technologies reshaping how companies grow.