AWS CEO calls AI business massive as Amazon capex tops $220 billion

AWS will keep pouring money into AI infrastructure because demand is outpacing what it can build, CEO Matt Garman said, as Amazon's 2026 capital spending reaches $220 billion.

"The potential AI business is just massive," Garman said in comments published Aug. 3, echoing Chief Executive Andy Jassy's warning that Amazon will still fall short of customer demand through 2027.

AWS revenue climbed 37 percent year over year to $42.2 billion in the second quarter, its fastest growth in 18 quarters, with operating income of $16.6 billion at a 39.4 percent margin. Amazon raised its 2026 capital expenditure forecast to about $220 billion, up $20 billion from prior guidance, citing higher memory chip costs. The company's contracted backlog jumped $132 billion in one quarter to $496 billion, with commitments stretching into 2028.

The spending puts Amazon at the center of a $725 billion hyperscaler investment cycle this year across Amazon, Alphabet, Meta and Microsoft, and the returns hinge on converting that buildout into durable earnings. Amazon shares rose about 9 percent in after-hours trading after the results.

Capacity Shortfall Runs Through 2027

Jassy told analysts on the July 30 earnings call that Amazon will not have enough capacity to meet all demand in 2026, and expects the same dynamic in 2027. "In fact, the demand we already have for 2028 is striking," he said. Amazon remains on pace to double its power capacity by the end of 2027 compared with 2025, with much of the planned 2027 capacity already reserved by customers.

The $496 billion backlog — work signed but not yet delivered — gives Amazon unusual visibility into future infrastructure needs. Cumulative multi-year commitments from major AI labs now exceed $225 billion, anchored by Anthropic's 10-year, $100 billion Trainium capacity commitment signed in April. The $20 billion increase in this year's capex forecast reflects higher memory costs rather than an expanded buildout footprint, Amazon said, with second-quarter capital expenditures alone totaling $54.2 billion, up 68 percent year over year.

The capacity constraint is not unique to Amazon. Alphabet has guided 2026 spending of $175 billion to $185 billion, Meta $125 billion to $145 billion and Microsoft $110 billion to $120 billion, bringing the four largest hyperscalers to roughly $725 billion — up 77 percent from 2025. Analysts at HyperFrame Research and Constellation Research said the industry remains in an infrastructure deployment phase, with compute, power and data center construction the binding constraints rather than customer interest.

Trainium Adoption Broadens

Amazon's custom silicon business and its AI business each independently crossed a $25 billion annualized run rate, growing at triple-digit rates year over year. Anthropic and OpenAI have made multi-year, multi-gigawatt commitments to Trainium, Amazon's in-house accelerator that competes with Nvidia's H100 and B200 data center GPUs. Matt Kimball, vice president at Moor Insights & Strategy, called the commitments "about the strongest endorsement a piece of silicon can get."

Jassy said Amazon is exploring selling Trainium chips outside AWS after interest from customers wanting to deploy the processors in third-party data centers. "We're actively having those conversations and exploring, and I expect there's a real chance we'll do that in the future," he said. A merchant silicon push would require public roadmaps, field engineering and channel support — go-to-market motions Nvidia has spent decades building.

The bet is that heavy upfront spending converts into durable free cash flow. Jassy separated the investment into long-lived data centers, which can remain productive for more than 30 years, and shorter-lived servers and networking gear that reach break-even in less than three years. He said AWS could "very possibly" become a trillion-dollar annual revenue business. Amazon trades at a premium to peers on the strength of that narrative, but the $220 billion capex plan — the largest single-year commitment among hyperscalers — leaves little room for demand to soften before returns materialize.

This article is for informational purposes only and does not constitute investment advice.