Amazon's $200B Quarter: Ads, AWS, and the Anthropic Gain

Amazon's Q2 2026 earnings, reported July 30, cleared a threshold no Amazon quarter has before: $200.6B in net sales, up 20% year over year. AWS grew 37% — its fastest rate in 18 quarters — advertising grew 26% to $19.8B, and net income printed at $62.6B. That last number is where the discipline starts.

Of the $80.9B in pre-tax income Amazon reported, $53.4B — roughly 66% — is a non-operating, pre-tax mark-to-market gain that the company says came "primarily from our investments in Anthropic." It is not cash, it is not operating profit, and it makes the headline net-income figure incomparable to any prior quarter. Most coverage reported the $62.6B and the $53.4B as two separate facts; almost nobody built the bridge between them.

This post builds that bridge, traces how the same Anthropic dollar shows up in as many as four different places in Amazon's disclosures, separates the audited ad-revenue number from Amazon's own unaudited attribution claims, and closes with what the quarter actually means for teams spending money on Amazon's ad and cloud platforms.

Start with the numbers nobody disputes. Total net sales of $200.606B grew 20% year over year from $167.702B in Q2 2025 — the first time Amazon has cleared $200B in a single quarter. Operating income rose 43% YoY to $27.461B, lifting the consolidated operating margin to 13.7% of worldwide net sales from 11.4% a year earlier. Every segment participated: North America grew 16% to $116.177B with segment margin up to 7.9% of segment revenue (from 7.5%), International grew 15% to $42.197B at a flat 4.1% margin, and AWS grew 37% to $42.232B.

The acceleration is the part worth underlining. On Amazon's own supplemental tables, consolidated net-sales growth excluding currency effects stepped up from 15% in Q1 2026 to 20% in Q2, and AWS stepped up from 28% to 37% on the same ex-FX basis. This was not a quarter flattered by comparisons — both the consolidated line and AWS sped up sequentially. Amazon does not publish the same ex-FX sequential comparison for North America or International, so treat the acceleration as consolidated-and-AWS, not segment-by-segment.

Net income was $62.647B, or $5.75 in diluted EPS, against $18.164B and $1.68 in Q2 2025 — a 245% increase in net income and 242% in EPS. Neither number survives contact with the income statement's next line. "Other income (expense), net" swung from $1.117B a year ago to $53.415B, which Amazon says came "primarily from our investments in Anthropic." That is a pre-tax, non-operating, mark-to-market gain on an equity stake — paper appreciation, not cash, and not operating profit. Measured against the quarter's $80.857B in pre-tax income, it is roughly 66% of the total.

The balance sheet tells the same story from the other side: total assets jumped to $1.096 trillion at June 30, 2026 from $818.0B at December 31, 2025, with "Other assets" ballooning from $122.6B to $284.1B — consistent with the Anthropic stake being marked up — alongside property-and-equipment growth to $446.0B from $357.0B.

So what did the operating business actually earn? Amazon does not publish that bridge, and neither did the press coverage we reviewed, which reported the $62.6B and the $53.4B as separate facts in separate paragraphs. Here is our reconciliation — explicitly labeled as our own estimate wherever we apply tax effects, because Amazon discloses only a blended provision ($18.199B on $80.857B pre-tax, roughly a 22.5% effective rate), not a per-item rate for the Anthropic gain.

Reported figures from Amazon's Q2 2026 Consolidated Statements of Operations. The ~22.5% tax-effecting rate is the quarter's blended effective rate ($18.199B provision ÷ $80.857B pre-tax income); Amazon does not disclose the tax treatment of the Anthropic gain specifically, so the after-tax adjustment and the ~$20.8B result are Digital Applied estimates, not company-stated metrics.

Read that bottom line carefully: roughly $20.8B of underlying net income on our estimate, up about 14% from Q2 2025's reported $18.2B. That comparison is deliberately conservative — Q2 2025's reported figure itself included $1.117B of pre-tax other income, so on a like-for-like basis (roughly $17.3B underlying a year ago) the increase is nearer 20%. Either way it is a genuinely strong quarter — operating income up 43% is nothing to explain away — but it is a very different story from "profit more than tripled." Any analysis, board deck, or competitor comparison that uses Amazon's $62.6B or $5.75 EPS without backing out the Anthropic mark is comparing apples to a balance-sheet event.

AWS net sales of $42.232B grew 37% year over year from $30.873B — Amazon's fastest AWS growth in 18 quarters, and a beat against the $40.54B StreetAccount consensus CNBC cited. AWS operating income rose 64% YoY to $16.621B, taking segment operating margin to 39.4% of AWS revenue from 32.9% a year earlier. The press release frames the quarter as a $169B annualized revenue run rate — remember that is quarterly-annualized math, not the $148.4B in trailing twelve-month AWS revenue.

The margin number needs its footnote. On the earnings call, CFO Brian Olsavsky disclosed that the 39.4% includes roughly $600M of mark-to-market gains on energy derivative contracts — a first-time-material item, per The Register's reporting. On Olsavsky's own framing, AWS margin expanded about 650 basis points YoY including that gain, or about 520 basis points excluding it — and the release's Q3 guidance explicitly assumes "no impact from energy derivative contract remeasurements" going forward. The underlying margin expansion is real; the headline overstates it by roughly a point and a half.

Demand-side disclosures were unambiguous. Jassy said on the call that AWS's contracted, not-yet-recognized backlog reached $496B in Q2, up $132B in the quarter alone, per The Register — the same quarter in which Anthropic's ten-year, $100B Trainium commitment from April 2026 was signed. The release adds that both Anthropic and OpenAI have made "multi-year, multi-gigawatt" Trainium capacity commitments. Coverage of the call by Benzinga and others put cumulative Trainium revenue commitments above $225B across customers — a press-aggregated figure from call commentary that does not appear in the written release, so treat it as reported, not company-published.

Two platform notes round out the AWS picture. Bedrock added more than ten fully managed foundation models in the quarter — including OpenAI's GPT-5.6, Anthropic's Claude Opus 5, Google DeepMind's Gemma 4, and SpaceXAI's Grok 4.3 — and Amazon says "hundreds of thousands" of customers now use Bedrock, with Q2 Bedrock spend exceeding all prior quarters combined (an unquantified vendor claim). Graviton5 reached general availability, with Amazon claiming up to 25% better compute performance than Graviton4; separately, the company says the Graviton family as a whole is now used by 98% of the top 1,000 EC2 customers — again the company's own figures.

One forward-looking wrinkle from the call: asked by Morgan Stanley's Brian Nowak whether Amazon might sell Trainium chips to third parties, Jassy said customers are increasingly interested in getting Trainium "separate from our cloud," that Amazon is "actively having those conversations," and that "there's a real chance we'll do that in the future," per The Register. If that happens, the "chips business" label in the next section becomes literal — today it is not.

Here is the structural pattern no single headline captured. Amazon holds a large equity stake in Anthropic. Anthropic committed — ten years, $100B, April 2026 — to run its models on AWS Trainium chips. That one relationship then surfaces in Amazon's Q2 disclosures in as many as four places: as AWS segment revenue, as part of the more-than-$25B "AI business" run rate, as part of the more-than-$25B "chips business" run rate, and — entirely separately — as the $53.4B unrealized equity gain that inflated net income. Same underlying dollar, several labels.

Synthesis by Digital Applied from Amazon's Q2 2026 press release (segment and run-rate disclosures) and The Register's earnings-call reporting (Corey Quinn, July 31, 2026). Amazon does not disclose Anthropic's share of any revenue line; the table maps where the relationship touches each disclosure, not dollar attributions.

The Register's Corey Quinn put the sharpest edge on the "chips business" framing — the more-than-$25B run rate is EC2 and Bedrock instance-rental revenue running on Trainium hardware, not chip sales, since Amazon has never sold a discrete Trainium chip to anyone.

Advertising services revenue reached roughly $19.8B in Q2 2026, up 26% year over year — $19.81B against the $19.43B StreetAccount consensus, per CNBC. That growth rate and the beat are the audited facts. Jassy's release quote kept it simple: "Advertising had another strong quarter with 26% year-over-year growth. There's a lot to be excited about, and we have much more coming for customers in the second half of the year and beyond." And on composition: "Sponsored Products continues to be our largest offering and a key driver of growth."

Amazon names three drivers in its own materials — AI-powered ad tooling, conversational shopping, and Prime Video's live-sports inventory. What it does not do is decompose how much of the 26% each driver contributed, and every engagement statistic attached to them is the company's own unaudited comparison. Treat the grid below as Amazon's story, clearly labeled.

For advertisers, the practical read is about auction dynamics, not press-release stats. A platform growing ad revenue 26% YoY while pushing AI-assisted campaign setup into eleven more countries is a platform where more of the budget-setting is happening inside Amazon's own optimization loop — the same pattern we unpacked in Alphabet's Q2 report the week before, where AI tooling took attribution credit for an ad beat. Independent measurement is the counterweight: if Ads Agent is setting your bids, your incrementality testing had better not be Amazon's dashboard. That is the core of how we run paid media engagements on retail media accounts — vendor automation where it earns its keep, measured against baselines the vendor does not control.

Sellers feel this quarter from the cost side too: ad-auction competition on a platform growing 26% compounds with the fee structure sellers have already been adjusting to in 2026. Rising retail-media efficiency claims from the platform do not automatically show up as rising net margin for the merchant.

Amazon raised full-year 2026 capex guidance to $220B, up from the $200B it guided in February and reaffirmed in April — Jassy attributed the increase to rising memory-chip prices, per CNBC's reporting from the call. The quarter itself saw $54.208B of purchases of property and equipment, up roughly 68% from $32.183B in Q2 2025, taking trailing-twelve-month capex (net of proceeds) to $169.0B, up 64% YoY.

The cash-flow consequence landed exactly where you would expect: trailing-twelve-month free cash flow flipped to an outflow of $7.604B, from an inflow of $18.184B a year earlier — a swing of roughly $25.8B that Amazon attributes primarily to a $66.1B YoY increase in property-and-equipment purchases tied to AI infrastructure. A company posting record revenue and a $62.6B net-income headline is, on a trailing cash basis, spending more than it generates. How Amazon's raised guide stacks up against the rest of the week's reports is the subject of our AI capex scorecard for July's earnings week.

On the call, Jassy described the AI demand curve as "very barbelled," per The Register — AI labs consuming enormous compute at one end, enterprises doing cost-avoidance at the other, and a middle of enterprise production inference "most of which aren't" yet running at scale, whose trajectory he would not promise matches the labs' curve. That is the honest tension under the $220B: the labs' end of the barbell is contracted; the middle is the bet.

Guidance was the quarter's one soft note. Amazon guided Q3 2026 net sales to $197.0B–$202.0B, or 9–12% YoY growth — below the roughly $204.1B LSEG analyst consensus CNBC cited — with operating income of $22.5B–$26.5B against $17.4B in Q3 2025. The release flags two distortions in the growth-rate comparison: roughly 400 basis points of drag from Prime Day timing (Prime Day landed inside the June quarter in both comparison windows) and an unfavorable foreign-exchange impact of roughly 80 basis points. Notably, the guidance also assumes "no impact from energy derivative contract remeasurements" — the item that flattered AWS's Q2 margin.

The market bought the quarter anyway — though how enthusiastically depends on which outlet and which snapshot you read. CNBC reported the stock "shot up more than 10%" in extended trading on the evening of July 30; Yahoo Finance's Stocktwits coverage headlined a roughly 7% after-hours gain. The primary press release states no stock figure at all — after-hours percentages are always secondary, snapshot-dependent market reporting. The defensible claim: a strong positive reaction, somewhere in the +7% to more-than-10% range in extended trading, varying by outlet and time stamp.

Four reads for teams whose budgets touch Amazon — as advertisers, sellers, or AWS customers.

There is also a build-vs-buy echo in the Ads Agent story. Amazon is making the case that its in-platform AI should run your campaigns; the counter-case is tooling you own, measured independently. It is the same calculus we laid out for agentic CRM buy-vs-build decisions: vendor agents optimize for the vendor's objective function first. For ecommerce operators weighing how much of their stack to hand to the platform, our ecommerce engagements start from exactly that ownership question.

Looking forward, two things are worth watching between now and the Q3 print. First, whether Amazon separates — or is pushed by analysts to separate — the overlapping "AI" and "chips" run rates into something auditable; Jassy's openness to selling Trainium "separate from our cloud" suggests the chips label may eventually earn its name. Second, the Anthropic mark itself: a stake large enough to add $53.4B of pre-tax income in one quarter is large enough to subtract meaningfully in another. The flywheel is real; it is also now a source of earnings volatility Amazon has never carried at this scale.