AMD Data Center Revenue +107% (Q2 2026): Record $11.5bn, a $13bn Guide — and the Stock Fell More Than 8%

AMD reported second-quarter results after the US close on Tuesday 4 August 2026 and beat on every line that was published in advance. Revenue was a record $11.536 billion, up 50% year-on-year and marginally above the top of the company's own $11.2 billion ±$300 million band. Data Center revenue more than doubled, up 107% to $6.718 billion and 58% of the company. Non-GAAP earnings per share were $1.66 against consensus near $1.62, gross margin hit the guided 56%, and third-quarter revenue was guided to approximately $13 billion — above the roughly $12.5 billion analysts had pencilled in. The shares fell more than 8% in after-hours trading. That gap between a clean beat and a negative reaction is the whole lesson of this print, and it is a mechanical one.

The preview of this report asked a narrow question: is the second-half accelerator ramp still dated where the company said it was, and what does AMD earn while it has not yet started? Both are now answered. What was not on the list — and turned out to be the more instructive part — is what happens when a company clears every visible bar and the price falls anyway.

What actually happened

Everything below is from AMD's own release (primary source: AMD Investor Relations, Q2 2026 results, published 4 August at 4:15 p.m. Eastern).

Two things in that table deserve to be read slowly. The revenue figure did not merely land inside the guided range — it cleared the top of it, which the preview treated as the less likely branch. And the third-quarter guide, at roughly 41% year-on-year growth and 13% sequential, is the company telling the market that the second-half step it has been describing since February is now in the numbers it is willing to commit to.

The number that did the work

Data Center revenue of $6.718 billion, up 107% from roughly $3.2 billion a year earlier, is the quarter. It is 58% of the company, and the CFO said so explicitly: Jean Hu's statement noted revenue rose 50% "driven by continued strength in our Data Center business, which represented 58% of company revenue in the quarter," and added that AMD expects Data Center sales "to accelerate in the second half of 2026."

The composition matters more than the total. Server processors are the mature, shipping side of that segment, and they are what got the revenue line above the guided band this quarter — accelerator revenue at scale was never in it. Lisa Su's statement puts the forward piece plainly: AMD enters the second half "with strong momentum as EPYC demand accelerates, Instinct deployments scale and Helios begins to ramp."

That last clause is the sentence the preview was waiting for. The second-half timing for the first gigawatt of shipments is intact on the company's own account, and the named list of Helios deployments in the release is now wider than the Meta agreement alone — Anthropic, Meta, Microsoft, OpenAI and Oracle. In July the company announced a partnership with Anthropic covering up to 2 gigawatts of MI450-series GPUs in Helios racks, on top of the 6-gigawatt Meta arrangement disclosed in February.

Confirmed is not the same as delivered, and the distinction is the honest one to hold. A press release restating a schedule is evidence about intent and readiness. The first hard checkpoint is second-half shipments actually being recognised as revenue, and no quarter reported before then can settle it.

Beat on every line, down more than 8%: the mechanism

This is the part worth understanding, because it recurs every earnings season and it is routinely explained badly.

A share price is not a bet on whether a company will beat its published consensus. It is a claim on a path of future cash flows, and by the time a report arrives that path already embeds an expectation — one assembled from the company's own guidance, from the run of recent results, from the disclosed size of the second-half commitments, and from whatever the market has decided those commitments are worth. Published analyst estimates are a visible approximation of that expectation. They are not the expectation itself, and after a very large run into a print the two can be some distance apart.

So a result can be a beat against the number on screen and a disappointment against the number in the price. Revenue $250 million above consensus and a guide $500 million above consensus are real, but they are increments; what a steep run-up requires is a change in the slope, and a quarter that executes the existing plan well does not supply one. Nothing in the release contradicted the plan. Nothing accelerated it either.

One caution on the figure itself. The 8% is an after-hours quote taken in the thin session immediately following the release, before the conference call had finished and on a fraction of normal volume. After-hours prints move, and the confirmed reaction is Wednesday's cash open — a session that also carries the US ISM services survey and the ADP employment report, so two unrelated inputs will be competing for the same tape.

The two EPS numbers, and why the gap narrowed

Anyone comparing headlines will encounter $1.66 in one place and $1.38 in another. Both are real, and the preview flagged the trap: match a reported figure to an estimate built on the same basis, or don't compare them.

What changed this quarter is the size of the wedge. In the first quarter, GAAP diluted EPS of $0.84 against non-GAAP of $1.37 was a gap of $0.53 — mostly amortisation of acquisition-related intangible assets, a non-cash charge following from past acquisitions, plus stock-based compensation, which is a real cost to existing owners even though no cash leaves the business. This quarter the same comparison is $1.38 against $1.66, a gap of only $0.28.

That narrowing is not evidence that the amortisation and compensation charges shrank. Look at the operating line: GAAP net income of $2.297 billion was larger than GAAP operating income of $1.990 billion, which is only arithmetically possible when items below the operating line contribute more than tax takes away. The composition of those items sits in the quarterly filing rather than the press release, and it is worth reading before treating a narrower wedge as improved operations. Non-GAAP operating income, by contrast, was $3.094 billion — above non-GAAP net income, which is the ordinary relationship.

The ramp shows up in the cash flow first

Free cash flow was $1.558 billion, against the record $2.6 billion reported three months earlier, while revenue rose about 13% sequentially. AMD also repurchased no shares in the quarter; the $221 million of buybacks disclosed for the first half all belong to the first quarter.

Read those two facts together and they describe the ordinary mechanics of a ramp. Inventory has to be built, capacity has to be committed and suppliers have to be paid in the quarters before the product ships and the revenue is recognised. Cash leaves early; revenue arrives later. A growth quarter that consumed cash relative to its predecessor is what the second-half schedule looks like from the balance sheet's point of view, and preserving cash rather than retiring shares is consistent with funding it.

None of that is a judgement, and quarterly cash flow is lumpy enough that a single comparison is not a trend. It is simply the counterweight to a revenue line growing 50%: the growth has a funding cost, the cost lands first, and both belong in the same picture. If the second half delivers, this reverses. If it slips, the cash went out anyway.

How this reaches an index you trade — and how much of it doesn't

Do the arithmetic openly, because it is the antidote to the "AMD moved the market" headline.

A market capitalisation on the order of $900 billion against an S&P 500 aggregate near $67 trillion is a weight of roughly 1.3%. A capitalisation-weighted index gives each constituent influence proportional to its size, so a 10% single-day move contributes about 0.13 percentage points to that index directly — noticeable, and nowhere near a day's typical range. In the Nasdaq-100 the same move lands considerably harder, because that index is a fraction of the S&P 500's aggregate size and excludes financials entirely, concentrating weight in exactly this kind of company. Treat both figures as approximate: index weights are float-adjusted, and the denominator is a quarter-end snapshot.

The larger channel is not weight at all. It is that a demand read from one accelerator vendor is immediately applied to everyone attached to the same spending cycle — the other chip designers, the foundries and equipment makers, the memory suppliers, the networking and power businesses, and the hyperscalers whose capital budgets are the demand. That is why a single report can produce a sector-wide session while the reporting company's own index contribution stays small. The same effect ran in the other direction after Microsoft's fiscal fourth quarter, when capital-spending guidance repriced its suppliers alongside its own shares — set out in the note on that record session. And the supply side has its own arithmetic: the memory shortage that split Apple and Amazon's July results is an input cost for everyone building accelerators.

The Gaming segment falling 31% while Data Center doubled is the same concentration seen from the inside. A company whose revenue is 58% one segment is, for practical purposes, a position on that segment's demand — which is why the read-through from this report to the wider complex is treated as more informative than one company's quarter has any right to be.

What would change the picture

Three things, none of them predictions.

The first is 26 August, when Nvidia reports. Until then AMD's demand commentary is the market's working estimate for the larger supplier, and that read-through is both a genuine information channel and an unreliable one: the two firms sell into overlapping customers with different product cycles, supply constraints and China exposure. A Nvidia report that corroborates accelerating second-half demand makes this quarter look like early confirmation; one that does not makes it look like share gain into a slowing market. Same AMD numbers, different meaning.

The second is the margin path through the ramp. Non-GAAP gross margin of 56% met the guide and the third quarter is guided to the same level — which is the more interesting claim, because early-ramp hardware is normally less profitable than mature product. Holding 56% while Helios ships would be evidence of pricing power in accelerators. Guiding it and then missing it would say the company is buying share in a market where someone else sets the terms.

The third is China, and the honest answer is that this release said nothing about it. AMD has previously said it is not building incremental China AI revenue into its outlook while export licensing for advanced accelerators is unresolved. Because the number is essentially absent from the plan, movement in either direction is asymmetric: it is upside not in the model rather than downside that is. It also remains a reminder that a chip company's revenue is partly a function of export control, which is a policy variable on an unpredictable clock.

What has not changed is that this is not a currency event. Exchange rates are set by interest-rate paths, growth, inflation, terms of trade, positioning and the global risk regime, and one company's quarterly revenue is an input to none of them. The connection that does exist runs through the discount rate: businesses whose revenue arrives years after the capital is spent are more sensitive to the level of interest rates than near-term earners are, and the policy rate setting that discount is a genuine dollar input, scored through the interest-rate factor on the USD currency page. Our coverage of the July Federal Reserve decision takes that side of the ledger.

The takeaway

The report answered its own question. The second-half ramp is dated where the company said it was, the server business is strong enough to clear the top of a guided band on its own, Data Center more than doubled to 58% of the company, and management committed to roughly 41% growth in the quarter now running. On the published bars, this was a good quarter followed by a better guide.

The shares fell more than 8% in the hours after it, and both facts are true at once. The resolution is not that the market was wrong or that the numbers were misleading — it is that a price already contained an expectation steeper than the published one, and executing a known plan well does not change a slope. What would change it is delivery: Helios racks recognised as revenue, 56% margin held while they ship, and cash flow turning back up once the ramp is funded rather than being funded. Those are second-half facts. The 26 August read from the larger supplier arrives first.

To learn how Pip Theory builds its fundamental scores, see the methodology overview.

Educational macro context only — not investment advice.