AMD Posts Record Data Center Quarter, but SpaceX's Nvidia Commitment Overshadows the Numbers

AMD's Q2 data center revenue more than doubled year-over-year to roughly $6.7 billion, yet a 7% share decline on record results shows that competitive anxieties — crystallized by SpaceX choosing Nvidia exclusively — have outpaced even exceptional execution.

AMD reported second-quarter revenue of $11.5 billion on August 5, a 50% year-over-year increase and the company's strongest quarter on record. Data center, now representing 58% of total sales, more than doubled from the year-earlier period to approximately $6.7 billion — a figure that would have seemed implausible five years ago for a company that spent much of the 2010s as a distant runner-up to Intel in x86 and an afterthought in accelerators. The result is a factual vindication of a strategic pivot that began in earnest when Lisa Su took the CEO chair in 2014 and accelerated sharply with the 2022 Xilinx integration and the subsequent MI300 ramp. Yet AMD's shares fell 7% to close at $482.05 on August 6, a reaction that encapsulates one of the central tensions in semiconductor investing: exceptional results can still disappoint a market that has already priced in perfection.

The proximate catalyst for the selloff was not in AMD's numbers but in a simultaneous announcement from SpaceX, which confirmed it has selected Nvidia as its exclusive AI compute partner — committing to H200, B200, and the forthcoming Rubin series for both orbital and ground-based infrastructure, including its newly unveiled Starmind satellite data center concept. Elon Musk's blunt characterization — "We will use only Nvidia chips" — landed like a verdict in a market already hyperattuned to any signal about competitive positioning. That a single customer decision by a single private company could move AMD shares by nearly a tenth underscores how much of the stock's valuation rests on the narrative of AMD as a credible Nvidia alternative, and how fragile that narrative can feel when a high-profile customer explicitly rejects it. The episode is a reminder that AMD's competitive story is as much a sentiment story as a fundamental one.

AMD has, however, been making moves that go well beyond chip-level competition. On July 29, AMD and Core Scientific announced a 15-year infrastructure partnership valued at up to $14 billion, giving AMD access to an initial 529 megawatts of dedicated AI data center capacity across five U.S. sites, with options to expand the relationship toward 2.5 gigawatts. Bernstein analysts estimated the arrangement could generate $14 billion in revenue for Core Scientific over its life; AMD also received warrants for 30 million CORZ shares. The logic of the deal — AMD securing physical compute capacity rather than simply selling chips into open markets — reflects a deliberate strategic reorientation. Rather than competing on silicon specifications alone, AMD is positioning itself as a vertically integrated capacity provider, a model closer to what Nvidia has pursued through NVL72 rack systems and cloud partnerships. The same week, AMD unveiled its Helios rack-level architecture, integrating server hardware and AI data center infrastructure into a unified design, and Microsoft's CEO confirmed that Azure plans to deploy both AMD Helios and Nvidia Vera Rubin in production AI workloads — a meaningful validation from one of the largest hyperscale buyers.

The hyperscaler picture is more nuanced than the SpaceX headline suggests. One report from Analytics India Magazine claimed AMD has secured significant compute orders from OpenAI, Meta, and Anthropic, breaking Nvidia's hyperscale dominance; those details have not been independently confirmed by AMD or the customers named, and should be treated with appropriate caution until corroborated. What is more firmly established is Reuters reporting that AMD is forecasting strong forward revenue growth driven by AI data center demand. On the software side, AMD launched ROCm.ai — a developer-experience platform featuring AI coding agents designed to reduce friction on AMD's GPU stack — representing an attempt to erode the moat that has long insulated Nvidia's CUDA ecosystem. AMD also unveiled a new GPU architecture intended to challenge Nvidia's forthcoming Rubin flagship, though performance benchmark details remain limited. Whether ROCm.ai can meaningfully shift developer behavior in practice remains an open question; the history of attempts to dislodge CUDA is long and largely unsuccessful.

AMD's capital expenditure trajectory offers context for the scale of this ambition. From a trough of $84 million in full-year 2013, the company grew annual capex to $974 million in fiscal 2025 — a twelve-fold increase over twelve years — though at a 3% intensity ratio against $34.6 billion in revenue, it remains lean by the standards of the industry it is trying to challenge. The $389 million spent in Q1 2026 alone suggests the pace is accelerating. Yet AMD's infrastructure pivot introduces a form of financial exposure more akin to a cloud provider than a chip designer. If AI infrastructure demand cools, or if AMD's GPU allocations within Core Scientific's facilities prove harder to fill than projected, the long-duration obligations embedded in a 15-year contract could become a liability rather than a structural advantage. The company is, in effect, making a leveraged bet on the durability of enterprise AI capital spending.

The picture that emerges from AMD's quarter is one of real but contested progress. Three signals warrant close attention in the quarters ahead. First, whether hyperscale customers such as Microsoft explicitly disclose AMD Helios utilization rates alongside Nvidia deployments — which would indicate whether AMD is winning additive share or filling secondary workloads. Second, the trajectory of ROCm developer adoption, a lagging indicator of any meaningful erosion in CUDA's software moat. Third, AMD's gross margin performance as the infrastructure-leasing model scales, since capacity commitments may carry a structurally different cost profile than direct chip sales. AMD has delivered numbers that confirm its trajectory; the market's 7% decline on a record quarter reflects not doubt about execution, but persistent uncertainty about the ceiling — and about who, ultimately, gets to define the architecture of the AI buildout.