Nvidia Details Its Vera CPU as AMD Opens Advancing AI 2026: the AI Chip War Climbs Another Rung
On July 21, 2026, Nvidia published the full whitepaper for its new Vera CPU, built on its own Olympus cores, and the next day AMD opened its Advancing AI 2026 conference in San Francisco with EPYC Venice and the Helios rack, according to CNBC, Tom's Hardware and TechTimes.
Nvidia Details Its Vera CPU as AMD Opens Advancing AI 2026: the AI Chip War Climbs Another Rung
Nvidia published, on July 21, 2026, the full architectural whitepaper for Vera, its new AI data center CPU and the company's first custom CPU core design in years, according to reporting from Tom's Hardware and CNBC. The next day, July 22, AMD opened its Advancing AI 2026 conference in San Francisco, announcing the EPYC Venice processor and the Helios rack, according to TechTimes. The two nearly simultaneous moves show that the fight for AI infrastructure has stopped being just about the accelerator GPU and now includes the CPU orchestrating everything around it.
Inside Vera: Olympus Cores Built for Agentic AI
Vera is built around Nvidia's custom Olympus core, designed to prioritize single-thread performance and memory bandwidth over raw core density. The chip packs 88 Olympus cores and 176 simultaneous threads, with features such as value prediction and elimination of unnecessary memory movement. In the SPEC CPU 2026 benchmark, the SPECrate integer suite, Vera hit a base score of 925 against 898 for a dual-socket system running AMD's Epyc 9755, a roughly 3% margin achieved with fewer total threads, according to Tom's Hardware. The test ran on a reference system, since Vera is not yet broadly available.
AMD's Answer: Zen 6 Venice and the Helios Rack
Advancing AI 2026 opened with the announcement of EPYC Venice, the first x86 server CPU built on TSMC's 2 nanometer process, and the Helios rack, capable of delivering 31 TB of HBM4 memory and powered by the MI455X accelerator, the company's most advanced yet, according to TechTimes. CEO Lisa Su's keynote, set for the morning of July 23, is expected to detail Helios deployment timelines, ROCm software progress and a first preview of the MI500 series roadmap for 2027, with participation from partners including Meta, OpenAI, xAI, Oracle, Microsoft and Cohere.
Why the CPU War Matters as Much as the GPU War
Historically, Nvidia dominated AI acceleration with GPUs while leaving the host CPU to Intel and AMD's ecosystem. Building a custom CPU core signals Nvidia wants to control the full performance stack around agentic AI, where a CPU's single-thread performance matters as much as raw GPU power for orchestrating complex tasks. AMD, for its part, is betting on tight integration between CPU, GPU and memory within a single rack, with Helios working as a direct answer to that same question: who delivers more end-to-end performance, not just a stronger standalone chip.
What It Signals for Cloud Capacity Buyers
For SMBs and agencies that depend on cloud providers to run AI models in production, more real competition among hardware vendors tends, over the medium term, to pressure pricing and widen infrastructure options, still concentrated in Nvidia today. It is not immediate relief: both Vera and Helios remain in testing or early launch stages, with no confirmed broad availability. But the move reinforces a point already raised in this outlet's coverage of TSMC's record results: the AI computing capacity bottleneck remains far from resolved, and any sign of more competition in the hardware chain is worth watching closely.
What's Clear
The fight for AI infrastructure has stopped being about which company has the strongest GPU alone. With Nvidia fully entering the CPU race and AMD doubling down on integrated racks, the deciding factor becomes who delivers the best end-to-end performance for agentic workloads, which depend on CPU and GPU working together. For anyone planning AI infrastructure over the next few years, this rivalry is worth tracking as a direct indicator of future compute cost and availability.