Nvidia just proved it doesn't need to stay in its lane, and the x86 duopoly should be sweating.
The Summary
- Nvidia's Vera CPU beat AMD's EPYC 9655P in Linux kernel compilation at Hot Chips 2026, marking the GPU giant's first public CPU win against a tier-one rival
- The company expects CPU revenue to more than double by fiscal 28, signaling serious commitment beyond proof-of-concept hardware
- This isn't just about faster compiles. It's about vertically integrated AI infrastructure that doesn't need Intel or AMD at all.
The Signal
Nvidia unveiled Vera at Hot Chips 2026, and the benchmark that matters most to developers compiling kernel code, not synthetic workloads or marketing fluff showed the chip outperforming AMD's EPYC 9655P. Linux kernel compilation is the kind of real-world test that separates hype from hardware. If Vera can compile faster than EPYC, it's not a science project. It's a product.
The timing aligns with Nvidia's broader CPU ambitions. The company projects CPU revenue will more than double by fiscal 2028, a forecast that only makes sense if Vera and its successors aren't niche products for AI labs. Nvidia is targeting the data center, the same market Intel and AMD have split for two decades.
"Nvidia's advancements in agentic AI hardware could redefine competitive dynamics, challenging existing CPU leaders."
Here's what makes this different from past CPU challengers: Nvidia already owns the AI accelerator market. Every hyperscaler, every AI lab, every company building agents is already buying Nvidia GPUs. Now Nvidia is offering the CPU to go with them. The pitch isn't "switch to our CPU." It's "why are you still buying someone else's CPU when we can optimize the whole stack?"
The company frames this as infrastructure for agentic AI, which is smart positioning. Agents don't just need inference speed. They need memory bandwidth, low-latency interconnects, and tight GPU-CPU integration. Nvidia can optimize all of that in ways AMD and Intel can't, because they don't control the GPU side.
Key advantages Nvidia is betting on:
- Vertical integration: GPU, CPU, and networking (Mellanox) all in-house
- Existing relationships: already inside every AI data center
- Workload alignment: designed specifically for agent inference, not legacy x86 compatibility
The Implication
If you're building AI infrastructure or buying compute for agent workloads, watch Nvidia's CPU roadmap. The Vera benchmark is an opening shot, not a one-off stunt. Doubling CPU revenue in two fiscal years means volume shipments, which means pricing pressure on AMD and Intel in the one market segment they still own.
For developers, this could mean fewer integration headaches and better performance per dollar if Nvidia can deliver on the promise of unified architecture. For Intel and AMD, it means defending their data center moat against an attacker who already has a foothold inside the building.