Amazon AWS Q2 2026: 37% Growth, a $25B AI Business, and a $53.4B Anthropic Mark That Dominates the Headline

AWS posted its fastest growth in 18 quarters, Amazon’s AI business crossed $25 billion annualized, and a $53.4 billion non-cash Anthropic mark inflated reported net income to more than twice operating profit — all in the same quarter.

What Happened

Amazon’s Q2 2026 earnings release shows AWS net sales growing 37% year over year to $42.2 billion — the segment’s fastest growth in 18 quarters — placing it on a $169 billion annualized revenue run rate. AWS operating income rose to $16.6 billion from $10.2 billion a year ago. Within that, Amazon separately disclosed that its AWS AI business has passed a $25 billion annual run rate and is growing at triple-digit percentages. Total company net sales rose 20% to $200.6 billion; total operating income rose 43% to $27.5 billion.

The headline net income number requires an immediate structural qualifier. Reported net income of $62.6 billion, or $5.75 per diluted share, is approximately 2.3 times operating income. The gap is almost entirely explained by a single line: a $53.4 billion non-operating, pre-tax gain that Amazon describes as “primarily from our investments in Anthropic.” That gain is non-cash, non-recurring, and set by private funding-round valuations rather than any public market. The honest operational number is $27.5 billion in operating income — strong in its own right, up 43% — but the reported EPS figure will be quoted in headlines without that context attached.

The spend side is equally structural. Purchases of property and equipment reached $54.2 billion in the quarter, up from $32.2 billion a year ago, with Amazon attributing the increase primarily to AI infrastructure investment. Trailing twelve-month operating cash flow stands at $161 billion, up 33% — but that capex trajectory compresses free cash flow materially, following the same pattern documented at Meta this quarter. On the demand side, Amazon announced that both Anthropic and OpenAI have made multi-year, multi-gigawatt commitments to its Trainium custom silicon — a significant statement about the training-layer competitive dynamic. Q3 guidance calls for net sales of $197–202 billion and operating income of $22.5–26.5 billion.

The Structural Read

Three threads from this quarter’s results deserve a structural frame, not just a financial one.

Returns: the AI demand question is being answered. AWS at 37% and Microsoft Azure at 43% in the same reporting week are both accelerating. That is not a coincidence and not a rounding effect. It is the clearest aggregate signal yet that the demand sitting beneath the AI infrastructure buildout is real and compounding, not projected. Amazon’s $25 billion-plus AWS AI run rate — disclosed separately, growing at triple-digit rates — is that demand made line-item explicit.

Earnings quality: the circular AI economy lands on income statements. The $53.4 billion Anthropic gain is the same structural phenomenon as Microsoft’s $3.2 billion Anthropic mark last week — just roughly 17 times larger. Amazon is simultaneously the investor in Anthropic, the cloud infrastructure vendor to Anthropic, and the financial beneficiary of Anthropic’s rising, IPO-trajectory valuation. As Anthropic’s funding-round marks climb, they flow directly into Amazon’s reported net income. This is the backstop economy fully realized: the hyperscalers provide the capital, the compute, and now record the paper appreciation — all from the same counterparty.

Training silicon: the CUDA decoupling reaches OpenAI. The announcement that both Anthropic and OpenAI have made multi-year, multi-gigawatt Trainium commitments advances the training-layer decoupling from Nvidia’s CUDA into new territory. OpenAI running training workloads on Amazon’s custom silicon — even as a disclosed commitment rather than a completed build — signals that the largest frontier labs are actively distributing their silicon dependency. The denominator here is gigawatts, not GPU count; the contracts are announced intentions, not fully realized infrastructure. But the direction is unambiguous.

Three Implications