SANTA CLARA, Calif. — Advanced Micro Devices on Tuesday reported the largest data-center quarter in its history and lifted full-year guidance, telling investors that its newest MI400-series accelerators were now shipping in volume to three of the four largest U.S. cloud providers and that its order book extended into the back half of 2027.

Revenue for the quarter ended March 28 came in at $9.84 billion, up 51 percent year over year and ahead of the $9.41 billion analyst consensus compiled by FactSet. Data-center segment revenue more than doubled to $5.62 billion, eclipsing the client and gaming businesses combined for the first time. Adjusted earnings per share landed at $1.74, eight cents above estimates. Shares rose roughly 7 percent in after-hours trading.

The print landed at the tail end of a megacap earnings cycle that has been defined by accelerating artificial intelligence capital spending. Hyperscalers including Microsoft, Alphabet, Amazon and Meta have collectively guided to more than $400 billion of capex this year, a figure that has reshaped expectations across the semiconductor supply chain even after last month’s collapse of the federal AI training moratorium and the opening of state-level fights in California and New York.

Chief Executive Lisa Su told analysts on a post-close call that customer demand for the MI400 line, which began commercial shipments in February, was “well in excess” of what AMD could supply through the third quarter, and that the company had “committed allocation through year-end across our top ten customers.”

“We are seeing a clear shift in customer behavior,” Su said. “Buyers who two years ago wanted a single-vendor stack are now actively engineering for multi-vendor accelerator pools. That works to our advantage.”

AMD lifted full-year revenue guidance to a range of $42 billion to $44 billion, from a prior range of $38 billion to $40 billion, and said it now expected data-center revenue to roughly triple year over year. Chief Financial Officer Jean Hu said gross margin in the quarter expanded to 54.1 percent and would push toward 56 percent by the fourth quarter as MI400 volumes climbed and inventory written down in 2024 cleared the books.

The results sharpened a question that has hovered over the AI hardware trade for most of the past year: how durable is Nvidia’s dominance once a credible second source reaches scale? Nvidia, which reported a blowout April 30 quarter of its own, still ships an estimated 88 percent of merchant AI accelerators by revenue, according to a tally published last week by the research firm Mercury Vector. But AMD’s data-center disclosures suggest the gap may be narrowing at the margins, particularly in inference-heavy workloads where MI400’s memory configuration has drawn favorable benchmarks.

“This was the quarter the bull case stopped being a story and started being a number,” said Vikram Doshi, a semiconductor analyst at Piper Sandler, in a note circulated to clients minutes after the release. “The MI400 is real, the backlog is real, and the customer concentration risk is finally easing because the hyperscalers want it to.”

Roughly 62 percent of the data-center segment came from three customers, AMD disclosed in its 10-Q. The company did not name them, but executives confirmed on the call that one of the largest U.S. hyperscalers had moved a “meaningful portion” of its 2027 procurement to MI400 and successor parts. People briefed on the contracts identified the customer as Microsoft, which has been retooling its Azure AI fleet around mixed-vendor pods since late last year. Microsoft declined to comment.

Su was repeatedly pressed on whether tightening grid constraints — driven both by raw demand and by the patchwork of state-level moratorium proposals now circulating — would cap upside. She conceded that “power availability is the single most-discussed topic in our customer reviews” but argued the constraint cut both ways for AMD, because customers facing power caps were under pressure to choose the most performance-per-watt option available.

“You are seeing customers redesign racks around power, not around any one vendor,” Su said. “That is a multi-source dynamic, by definition.”

The earnings arrived against a softer macro backdrop than at the start of the year. Brent crude settled near $90 a barrel Tuesday, down from its $125 mid-March wartime peak, after Doha mediators last week extended the Iran ceasefire framework by 90 days and Gulf supply normalized. The S&P 500 has recovered all of its war-period losses and is now slightly positive on the year, helped by precisely the kind of capex-driven hardware results AMD reported.

Inside the company, the quarter capped an 18-month repositioning that began with the late-2024 acquisition of an inference-tooling startup and accelerated with a series of design wins announced at the company’s October architecture day. AMD said its custom silicon business, which builds semi-custom accelerators for a small number of hyperscaler customers, had also grown sharply and was now expected to contribute more than $4 billion in 2026 revenue, up from a prior estimate near $2.5 billion.

Not every line was clean. Client and gaming revenue both came in slightly below consensus, with Hu citing softer PC demand in greater China and a transition quarter for the Ryzen desktop refresh. Embedded revenue, the legacy Xilinx business, fell 9 percent on weakness in industrial end markets.

Su said the company would host an investor day on June 17 in San Jose to detail its MI500 roadmap and a new co-packaged optics interconnect that AMD is developing with two unnamed foundry partners. She declined to comment on whether AMD had begun discussions with the California energy regulator over the state’s proposed data-center moratorium, saying only that the company was “engaged in many such conversations across many jurisdictions.”

Officials at the company said additional details on the MI400 supply ramp and 2027 commitments would be provided at the June event.