AMD data center sales double as shares fall despite an earnings beat
Under Lisa Su, AMD posted $11.54 billion in revenue and expects about $13 billion next quarter as Helios begins ramping.
By Ryan Merket · Published
Why it matters
AMD's AI challenge has moved beyond selling individual GPUs. Helios, EPYC and customer-linked warrants show the capital, systems engineering and commercial concessions required to loosen Nvidia's control of AI infrastructure.
AMD (@AMD) reported record second-quarter revenue on August 4th as data center sales more than doubled, yet its shares fell 7.63% in extended trading after a run that had nearly tripled the stock over the previous year, CNBC reported.
The Santa Clara, California-based semiconductor maker generated $11.54 billion in revenue during the quarter ended June 27th, up 50% from $7.69 billion a year earlier. Adjusted earnings reached $1.66 per share. Both figures exceeded LSEG consensus estimates cited by CNBC, which called for $11.28 billion in revenue and adjusted earnings of $1.62 per share.
Data Center revenue reached $6.7 billion, a 107% increase from the year-ago quarter, driven by demand for EPYC server processors and Instinct AI accelerators. The segment accounted for 58% of AMD's quarterly revenue, according to AMD's earnings release.
That mix marks the latest step in the reinvention led by Lisa Su, the electrical engineer who joined AMD in 2012 and became CEO in 2014. Su earned three electrical engineering degrees from MIT and spent 13 years at IBM before moving through Freescale Semiconductor to AMD. Her tenure has centered on the Zen CPU architecture, EPYC's push into servers and the attempt to build Instinct into a credible alternative to Nvidia's AI accelerators.
Data center has become AMD's center of gravity
AMD's results show how far its revenue base has shifted toward AI infrastructure. Client and Gaming revenue rose 6% to $3.8 billion, with a 23% increase in client processor sales offset by a 31% decline in gaming. Embedded revenue increased 19% to $977 million.
Data Center alone produced about 58 cents of every revenue dollar AMD booked during the quarter. That business includes the GPUs used to train and run AI models and the CPUs that coordinate workloads, feed accelerators and handle general-purpose computing inside AI clusters.
AMD reported GAAP net income of $2.3 billion, or $1.38 per diluted share, compared with $872 million, or 54 cents per share, a year earlier. The comparison requires context: AMD's second quarter of 2025 included $800 million in inventory and related charges tied to U.S. export controls on MI308 data center GPUs. Those charges depressed the prior year's margins and earnings, making the year-over-year profit expansion look steeper.
AMD expects third-quarter revenue of approximately $13 billion, plus or minus $300 million. The midpoint would represent 41% annual growth and a 13% sequential increase. It also exceeds the $12.52 billion LSEG consensus cited by CNBC.
The after-hours selloff points to an expectations gap rather than a weak quarter. CNBC reported that some analysts had been looking for guidance as high as $14 billion. AMD beat the published consensus and forecast another record quarter, but investors had already priced in a more aggressive acceleration after the stock's sharp rise.
Helios raises the stakes against Nvidia
The next test is whether AMD can convert signed infrastructure commitments into shipments and revenue at the pace investors expect.
AMD launched Helios on July 23rd as its first rack-scale AI system, combining 72 MI455X GPUs, 18 sixth-generation EPYC CPUs, Pensando networking and ROCm software. Selling complete racks gives AMD a more direct answer to Nvidia's integrated systems, where chips, networking and software are designed and sold as a coordinated platform.
AMD says Helios is in production and will begin ramping in the second half of 2026. OpenAI expects to bring Helios systems online beginning in the fourth quarter, while Meta has started testing and validating workloads on the racks.
Those customers have made unusually large commitments. Under an October 6th, 2025 agreement, OpenAI agreed to deploy up to six gigawatts of AMD GPUs across multiple generations, starting with a one-gigawatt MI450 deployment in the second half of 2026. AMD issued OpenAI a warrant for up to 160 million AMD shares, with vesting tied to deployment, commercial and share-price milestones.
Meta (@Meta) signed a separate six-gigawatt agreement on February 24th. Meta's first gigawatt is scheduled to use custom MI450-based GPUs, sixth-generation EPYC CPUs and Helios racks. AMD also issued Meta a performance-based warrant for up to 160 million shares.
The warrants reveal the cost of challenging Nvidia (@nvidia) in a market where software compatibility, supply certainty and deployment experience can matter as much as accelerator specifications. AMD is giving two marquee customers a direct economic interest in successful deployments while using their orders to validate Helios for other cloud providers and AI developers.
Chief Financial Officer Jean Hu said AMD expects Data Center sales to accelerate during the second half of 2026. Revenue already shows that EPYC and Instinct demand is reaching material scale. The share reaction sets a higher bar: AMD must now turn gigawatt agreements and a new rack architecture into quarter-by-quarter sales without slipping on production, software or deployment schedules.