On December 1, 2025, IBM questioned the economics of hyperscalers’ AI infrastructure plans. Using a hypothetical 100-gigawatt global buildout costing $8 trillion, he concluded: “There’s no way you’re going to get a return on that.”
The warning was not irrational: AI infrastructure is expensive, chips depreciate quickly, and not every investment—or every provider—will earn an adequate return. But Amazon’s latest results make the categorical conclusion increasingly difficult to defend.
In Q2 2026:
- AWS revenue rose 37% to $42.2 billion, its fastest growth in 18 quarters.
- AWS operating income increased 64% to $16.6 billion.
- AWS achieved a 39.4% operating margin.
- Amazon’s AI and custom-chip businesses each exceeded a $25 billion annual revenue run rate, with both growing at triple-digit rates.
Amazon is now planning approximately $220 billion of 2026 capital spending, yet says demand continues to exceed available capacity and much of its 2027 AWS capacity is already reserved. AWS’s contract backlog reportedly reached $496 billion.
Amazon is simultaneously increasing AI investment, accelerating AWS growth and expanding cloud operating profit.
The better conclusion is not that every dollar of hyperscaler spending will pay off. It is that the companies owning scarce AI capacity, customer demand and global distribution may earn substantial returns—while companies that avoided the investment risk becoming dependent on them.
IBM may ultimately prove correct that parts of the AI buildout are excessive. But Amazon’s results suggest that refusing to participate carries its own, potentially greater, cost.