SANTA CLARA – AMD’s Epyc server processors have been the most credible challenge to Intel’s decade-long grip on the data center. By the end of this year, AMD may have a bigger problem than Intel.
Nvidia shipped hundreds of thousands of Grace standalone servers and more than 2.5 million Grace CPU units by May 2026, and its newer Vera chip, the successor that Grace built toward, now posts benchmark results that outpace x86 processors from both AMD and Intel by 50 percent, according to Nvidia’s performance assessments. The next server CPU race, analysts say, may already have a winner.
The stakes are measured in hundreds of billions. Nvidia, whose GPU business transformed the data center over the past four years, now projects $20 billion in Vera server CPU revenue for the current fiscal year, a figure that places it ahead of AMD’s entire data center segment, which generated $5.8 billion in the first quarter of 2026. Intel’s data center and AI revenue reached $6.3 billion in the second quarter.
Counterpoint Research’s analysis concludes that ARM-based server CPUs could capture nearly 90 percent of the server CPU market by 2029, compared to approximately 13.2 percent at the end of 2025. That is not a gradual shift. It is a structural displacement, and Nvidia’s Vera is positioned at its center.
The Vera chip’s performance advantage stems from architecture as much as silicon. Where Intel’s Xeon and AMD’s Epyc processors are built around x86 instruction sets optimized over decades for general-purpose computing, Vera is engineered from the foundation for AI workloads, high-bandwidth memory access, and the inference throughput that hyperscalers prioritize when choosing what to build their next cluster generation on. OpenAI, Anthropic, and SpaceX are among the early Vera customers, and their purchasing decisions signal where broader data center demand follows. The chip also outperforms Nvidia’s own Grace processor by 1.5 times on the benchmarks the company has published, according to the Nvidia developer blog.
AMD’s current position is not without strengths. Its MI350 GPU line has attracted significant investment, including a $5 billion strategic commitment from Anthropic alongside a multi-year chip deployment agreement announced earlier this month. That deal secured AMD a seat at the AI infrastructure table. It did not settle the CPU question, which is a separate and growing competitive front.
Intel’s position is more precarious. The company’s shift toward AI-optimized silicon through its Gaudi accelerator line has not produced the market traction that Nvidia’s CUDA-powered ecosystem generates, and its foundry business complications have added execution risk to a turnaround that investors have been waiting on since 2022. Intel’s data center AI revenue growth, while positive in absolute terms, has lagged the pace set by competitors whose products are built natively for the AI stack rather than adapted to serve it.
AMD’s role as the preferred x86 challenger in AI data centers faced a stress test this month when the Nasdaq 100 correction exposed chip sector vulnerabilities over four consecutive sessions. The sell-off hit AI infrastructure stocks disproportionately and raised a specific question for AMD: can its GPU gains offset long-run erosion in CPU market share as ARM-based alternatives expand their footprint?
Nvidia’s answer to that question is implied in the Vera product line itself. By entering the CPU market, Nvidia is not selling an alternative to its own GPUs. It is selling an integrated compute platform, CPU, GPU, and interconnect together, where the entire stack is Nvidia-optimized. That integrated pitch is harder for AMD to replicate because AMD’s CPUs and GPUs are sold to different buyers inside many enterprise purchasing structures, and because Intel’s ecosystem advantages in enterprise software compatibility give AMD less room to maneuver than it has in pure AI workloads.
The Counterpoint forecast of 90 percent ARM market share by 2029 is a projection, not a guarantee. Market forecasts in semiconductor hardware have been wrong before, in both directions, and the x86 ecosystem’s installed base is large enough that migration timelines in enterprise are measured in years, not quarters. AMD and Intel maintain meaningful revenue from existing x86 customers who are not migrating infrastructure ahead of schedule.
What is not uncertain is the direction. The $20 billion Vera revenue projection for this fiscal year would represent a CPU market entry at a scale no new server architecture has achieved this quickly. If that projection holds, it places Nvidia’s CPU business in a single year ahead of either AMD’s or Intel’s comparable data center revenue, with Vera Rubin rack deployments already running at CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud, making the disruption less a future event than a present condition.
How AMD responds to that trajectory, whether through deeper integration of its GPU and CPU roadmaps, further AI partnership investments, or aggressive pricing in existing Epyc segments, is the question that no earnings release has yet answered. The next set of numbers, due in calls over the coming weeks, will be the first real measure of whether the Vera effect is showing up in AMD’s data center bookings. Intel, meanwhile, is still waiting for signs that its restructuring has produced a product cycle capable of competing at the pace the AI infrastructure market now demands.
