After several quarters of declining to quantify it, Amazon has put a number on AWS's AI business. The number is a run rate.
The segment result
AWS revenue was $42.2 billion, up 37% year on year, with operating income of $16.6 billion against $10.2 billion — a 39.4% margin. Andy Jassy described it as the fastest growth in 18 quarters. The release headline frames it as "a $169 billion annualized revenue run rate", which is simply the quarter multiplied by four.
The disclosure, and what it is not
Two bullets in the release state that Amazon has "exceeded a $25 billion annual revenue run rate" for the AWS AI business, and separately for its chips business, both "growing triple-digit percentages". This is the first AI revenue figure Amazon has given. It is also, by construction, unauditable: $25 billion annualised is roughly $6.25 billion in the actual quarter, and Amazon reports no quarterly AI revenue line against which anyone can check it. Nor is "AI business" defined anywhere — there is no disclosure of what it bundles, so it may include general compute sold to AI customers.
Do not add them together
The two $25 billion figures cover overlapping businesses. Trainium silicon sold into AI workloads sits in both narratives, so summing them to $50 billion double-counts. And a small note on the growth rate: Jassy's own quoted number is 36.7%, which the headline rounds to 37% — the "fastest in 18 quarters" superlative rides on that rounding.
Why disclose it this way
A run rate is a number designed to be quoted rather than checked. Microsoft has done the same with Azure AI, and the pattern is now standard across the hyperscalers: give the market a large, growing, triple-digit figure that never has to reconcile to a reported segment. Against that, the AWS numbers that are audited — $42.2 billion of revenue and a 39.4% operating margin while capex runs at $54.2 billion a quarter — tell a more useful story about whether the buildout is paying for itself.