Nvidia Vera CPU targets two-thirds of server market
Key insights
- Nvidia CFO confirmed the 4 million Vera CPU FY2027 target, elevating it from analyst estimate to company guidance.
- Vera's bundling with GB300 and Rubin systems embeds CPU adoption into GPU infrastructure decisions already made by hyperscalers.
- At ~$5,000 per unit and 4 million units, Nvidia's projected $20B CPU revenue would approach two-thirds of the entire $30B x86 server CPU market.
Why this matters
Any AI infrastructure team planning datacenter builds over the next 18 months is now evaluating a procurement landscape where CPU and GPU choices are no longer independent decisions, since Nvidia's bundle mechanics tie Vera adoption to GPU platform selection. For Intel and AMD, the threat isn't a better chip competing on specs alone but a distribution channel they cannot replicate without a comparable GPU ecosystem. Founders and technical leaders building on cloud infrastructure should expect hyperscaler pricing and architectural choices to reflect this CPU consolidation, particularly as Nvidia gains leverage across both compute layers simultaneously.
Summary
Nvidia is on track to become the world's largest server CPU supplier by revenue, with analyst modelling projecting 4 million Vera CPUs shipped in FY2027 at roughly $5,000 each, totaling approximately $20 billion and capturing close to two-thirds of a $30 billion x86 server CPU market currently owned by Intel and AMD.
The threat isn't purely speculative. Nvidia CFO Colette Kress has publicly confirmed the 4 million unit target is on track, converting analyst projections into guided commitments. Vera's claimed 1.5x per-core performance and 2x performance-per-watt edge over x86 give it structural advantages, but the harder competitive moat is bundling: GB300 and Rubin system configurations are designed such that new datacenter builds effectively require Vera, removing CPU selection as a free variable for hyperscaler procurement teams.
Essentially: (Nvidia) is using GPU platform lock-in as a distribution channel for a CPU that Intel and AMD have no comparable bundling lever to counter.
- 4 million units at ~$5,000 each equals ~$20B revenue, versus Intel and AMD splitting a shrinking remainder of a $30B market.
- GB300 and Rubin bundle mechanics mean Vera adoption is tied to GPU infrastructure decisions already locked in.
- CFO confirmation shifts this from sell-side modeling to guided company outlook.
If the projection holds, the server CPU market's competitive structure will look fundamentally different by 2027 than it did in 2024.
Potential risks and opportunities
Risks
- Intel faces accelerating revenue compression in its datacenter CPU segment if Vera bundling captures hyperscaler refresh cycles through 2026-2027, potentially triggering further credit rating pressure on top of existing foundry losses
- AMD's EPYC roadmap loses its primary selling environment if GB300 and Rubin system designs structurally exclude x86 CPU substitution, stranding partner OEMs who built supply chain relationships around EPYC server deployments
- Hyperscalers that locked in multi-year Nvidia GPU agreements before Vera bundling terms were disclosed may face higher total platform costs than modeled, with limited renegotiation leverage until contract renewal windows in 2027-2028
Opportunities
- ARM server CPU vendors (Ampere, Qualcomm's datacenter unit) gain a positioning window by offering x86-independent CPU options to hyperscalers looking to avoid full Nvidia platform lock-in
- Datacenter infrastructure consultancies and procurement advisors see increased demand from enterprises and cloud builders needing independent analysis of true bundling obligations in Nvidia system contracts
- Intel and AMD could accelerate co-marketing or joint reference architecture programs targeting mid-tier cloud providers who lack the negotiating scale to push back on Nvidia bundle terms and need a credible x86-only alternative stack
What we don't know yet
- Whether hyperscalers (AWS, Google, Microsoft) have negotiated contractual carve-outs that let them substitute x86 CPUs in GB300 or Rubin configurations, or whether bundling is truly mandatory
- What Vera's actual per-core performance benchmarks look like on real datacenter workloads versus Nvidia's claimed 1.5x advantage, since no independent third-party validation was cited in post-earnings analyst notes
- How Intel's datacenter roadmap through 2027 responds to guided Vera volume, given Intel has not publicly addressed Vera bundling as a specific competitive risk in recent earnings calls
Originally reported by tomshardware.com
Read the original article →Original headline: Analyst Says Nvidia On Track to Capture Two-Thirds of x86 Server CPU Market From Intel and AMD — 4 Million Vera CPUs Projected for FY2027 at $20B Revenue