Advanced Micro Devices closed Tuesday at $370.68, reported $10.25 billion in revenue and $1.37 in non-GAAP earnings per share, and opened Wednesday sixteen percent higher. [1] The numbers behind the move: revenue up 38% year-over-year from $7.44 billion; data center segment revenue up 57% to $5.78 billion versus a Street consensus of $5.56 billion; net income up 95% to $1.38 billion on a GAAP basis and to $2.27 billion on a non-GAAP basis; and a Q2 guide of approximately $11.2 billion, against an LSEG consensus of $10.52 billion. [1] [2] On the call, CEO Lisa Su said the company has "strong and increasing confidence" in its ability to reach "tens of billions of dollars in data center AI revenue next year and to exceed our long-term growth target of greater than 80 percent in the coming years." [1]
The print resolves the inflection question the paper staged Tuesday. The bar for membership in the AI-capex chip-vendor cohort was set at $10 billion in revenue, $5.5 billion in data-center, an explicit MI450 H2 ramp, and a Q2 guide that signaled accelerating not stabilizing growth. AMD cleared every gate. The data-center number, which a year ago was $3.67 billion, is now larger than the entire company's revenue was eight quarters ago. The Q2 guide implies sequential growth of 9% from a print that was already 38% above last year. Su's "tens of billions" line is the same vocabulary Jensen Huang has been using on Nvidia calls for six quarters; that AMD is now using it is the news.
The Meta deal is the operating chassis. Confirmed on the call: shipments to support the first one-gigawatt deployment of the multi-year, six-gigawatt agreement begin in the second half of 2026, powered by a custom AMD Instinct GPU based on the MI450 architecture, with sixth-gen EPYC "Venice" CPUs running ROCm software on the AMD Helios rack-scale platform. [3] The agreement, announced February 24, included a performance-based warrant for up to 160 million AMD shares — roughly 10% of the company — vesting on shipment milestones tied to AMD share-price thresholds up to $600. [3] What was a press release in February became, on Tuesday's call, an executable document with a calendar attached. Su's framing: "Together with our previously announced OpenAI partnership, these engagements position AMD as a core partner to the world's largest AI infrastructure builders." [1]
The cohort split the paper has been documenting since Palantir's eighty-five-percent print Monday is now harder to read as a single trade. Through last weekend, the binary was clean: the Magnificent Seven and the AI-capex cohort were spending $715 billion in 2026 versus the Berkshire-anchored quality-cyclical cohort spending $13 billion. AMD on Tuesday — and Micron, which crossed $700 billion in market cap the same day on solid-state-drive shipment news, and Intel, which had its best month in 25 years in April — slot a second tier between those two: the silicon vendors who absorb capex on the way down. They are not the spenders. They are the suppliers. The $715 billion does not move without them.
The structural piece is what AMD is doing that Nvidia is not. AMD is the only US chip vendor that builds both leading-edge GPUs and leading-edge x86 CPUs, and the inference-driven AI workloads that have become the dominant capex line item turn out to need both. CFO Jean Hu's opening line on the call — "AI is now the primary driver of our revenue and earnings growth" — was paired with the segment data: client-and-gaming revenue up 23% to $3.6 billion, embedded up 6% to $873 million. [2] Each of those segments contains CPUs sold into the agentic-AI build-out. AMD's recent collaboration with Intel on a new x86 instruction set called AI Compute Extensions, which the companies say boosts compute density by 16x, is the structural answer to a market that, eighteen months ago, looked like it was going to be a GPU-only build. [1] CPUs went from "secondary to GPU" to "complement to GPU" between Q4 2025 and Q1 2026. AMD is the only company on the public market that prints both lines.
What this print does not yet resolve is the supply side. Su's prepared remarks acknowledged "supply constraints" twice; the global memory crunch — which has pushed Micron's market cap up nearly 700% in twelve months — and the advanced packaging bottleneck at TSMC are the operational ceilings on AMD's H2 ramp. [1] Helios, AMD's first full rack-scale system designed to compete with Nvidia's Grace Blackwell and Vera Rubin platforms — both of which sell for upward of $3 million per rack — ships later this year. [1] OpenAI and Meta are signed up. The question on the analyst calls, repeated four times Tuesday night, was the yield curve on the MI450 silicon. Su's answer was that customer engagement was strengthening and that "leading customer forecasts" were "exceeding our initial expectations." [4] Translated: the demand is real; the capacity is not yet certain.
The Cerebras counterparty is the cleanest stress test. Cerebras Day Three — the AI silicon competitor whose IPO headline halved from $40 billion to $26.6 billion last week — has its filings reviewed against AMD's print on the same trading desks Wednesday morning. The earlier read was that Cerebras was a venue-rich/customer-poor IPO trying to thread a market that had just absorbed $715 billion in Mag-7 capex commitments. AMD's data-center beat does not necessarily compress the Cerebras book, but it raises the floor on what a chip-vendor IPO has to demonstrate to clear at the original price. The cohort is bigger than it was Monday. The new entrants have to be earned through.
Stock prices, as always, lag the structural change by a quarter. AMD shares were up 66% year-to-date heading into the print and up another 16% on Wednesday's open. The CFO's language about "leverage in our operating model as we invest for accelerated growth while expanding profitability" — the line on the press release every public-market analyst will quote — is what the cohort needs to repeat each quarter to keep the multiple where it is. [4] Q2's $11.2 billion guide implies that the operating leverage is just beginning to show in the gross margin (55% non-GAAP for Q1, expected to be 56% in Q2). [2] If Q3 prints at $12 billion-plus and gross margin moves to 57%, the discount to Nvidia narrows. If it doesn't, AMD becomes a story stock again rather than an earnings story.
The macro overhang Disney's CFO was hedging on a separate call — "if a significant further rise in fuel prices from current levels could eventually lead to changes in consumer behavior" — does not show up in AMD's order book. Hyperscalers buy on multi-year forward commits; their procurement is not gated by quarterly oil prices. What gates AMD is TSMC capacity, HBM4 memory supply (the partnership AMD announced with Samsung, fittingly, was disclosed on the same earnings slide), and the MI450 yield curve. [4] The Iran war, in this story, is a non-input. The chip-vendor cohort prints into a different macro than the parks-and-streaming one Josh D'Amaro is running.
The cohort is bigger. The capex is real. The ramp is calendared. The MI450's first gigawatt at Meta begins shipping in roughly nine weeks. AMD's stock, as of Wednesday's open, had nearly tripled in twelve months. Lisa Su's "tens of billions of dollars in data center AI revenue next year" is the line every analyst will fact-check against the next two prints. The first arrives August 5: Q2 actual versus an $11.2 billion guide. The second arrives in early November: a Q3 print into the H2 MI450 ramp. Between now and then, the structural change Tuesday's earnings document recorded — that AMD is now the second chip vendor in the room where AI capex is allocated — gets either confirmed or unwound on a quarterly cadence.
Eight months ago, AMD was Nvidia's distant runner-up. Tuesday, it became Nvidia's complement. The Mag-7 capex envelope just got a second beneficiary. That is the news.
-- THEO KAPLAN, San Francisco