The CPU market's newest competitive front is the benchmark itself. Nvidia, Arm and their rivals are increasingly making performance claims built on AI agent workloads and test methods the vendors helped define, not on the general-purpose suites that once settled processor arguments. That shift matters more than any single doubling claim, because it changes what a published number can tell a US buyer.
The Agentic Workload Becomes the Unit of Competition
Two of the recent stories follow the same template. Nvidia's Vera CPU delivers twice the performance of Intel and AMD chips in AI agent workloads, Modal claims, according to Tom's Hardware. Arm says its AGI CPU delivers nearly 2X the modeled rack performance of AMD's EPYC 9755 across seven agentic AI workloads, also per Tom's Hardware. Different companies, different silicon, nearly identical framing: an agentic AI workload, a comparison against AMD or Intel, and a figure near two times.
That convergence is the story. When competing vendors independently choose the same workload category and the same rough multiplier as the headline, the benchmark has become a marketing instrument as much as a measurement. Agentic workloads are real and growing, but they are also young enough that no settled, industry-wide test exists for them. Each vendor is effectively grading its own exam.
Methodology Is Now the Battleground
Arm's claim is the more revealing of the two, because it is explicitly built on a new evaluation methodology developed with Fujitsu, Meta and OpenAI, as Tom's Hardware reported. That pedigree gives the number weight. It also means the result depends on choices about workload mix, modeling assumptions and what counts as a rack. None of those details are resolved by a headline multiplier.
The honest reading is not that Arm's claim is false. It is that the claim is conditional. A nearly 2X modeled rack advantage across seven agentic workloads is a strong result inside a defined methodology. Whether it holds under a customer's own mix of inference, retrieval and orchestration work is a separate question, and one that US enterprise buyers will end up answering in their own evaluations rather than in press materials.
Nvidia's CPU Ambition Cuts Both Ways
For US technology companies, the Modal result is significant less for the multiplier than for what it signals about Nvidia. A Nvidia CPU winning an AI agent workload comparison against Intel and AMD, as reported by Tom's Hardware, suggests the company is extending its data center position beyond accelerators into the processor socket. Modal's apparent willingness to say so publicly gives the claim commercial weight.
That is a competitive problem for Intel and AMD in their home market, and it is a strategic question for US buyers who have spent years diversifying away from single-vendor dependence in AI infrastructure. A credible Nvidia CPU option expands choice at the system level while concentrating more of the stack in one supplier. Both effects are real, and neither is settled by a throughput figure from one deployment.
