Arm Holdings announced the AGI CPU on March 24 — its first production silicon in 35 years — and the announcement came with a $15 billion pitch. The market responded with a more skeptical answer.
Arm CEO Rene Haas projected that the new chip business would generate $15 billion in annual revenue by 2031, roughly 60 percent of the company's total revenue target that year. The company already has $1 billion in customer commitments for 2028, expected to double annually through 2030. The market responded the next trading day: Arm's stock surged 20 percent intraday on March 25 before closing up 16 percent, adding $23 billion in market value.
But apply a multiple of four times sales — broadly in line with the average of chipmaking peers Nvidia, AMD, Intel, and Broadcom for 2030 — and the market's valuation bump only bakes in an extra $6 billion of incremental revenue. That is less than half of the $15 billion goal. The market is telling Haas to dial back his optimism, even as it rewards the ambition.
The disconnect is the story.
The chip itself is real. Arm built a 136-core CPU on TSMC's 3nm process with Neoverse V3 cores, targeting 300W thermal design. Meta is the lead customer and co-developer. OpenAI, SAP, Cerebras, Cloudflare, F5, SK Telecom, and Rebellions are listed as early customers, with commercial systems from Lenovo, Supermicro, and ASRockRack. The customer list is notable: these are companies that also license Arm's architecture — meaning Arm is now selling finished silicon to businesses it has historically depended on as royalty clients.
That tension — licensor competing with licensees — is the part of this story that holds up. For 35 years, Arm operated as a neutral infrastructure provider. The AGI CPU ends that positioning. Broadcom, Marvell, Nvidia, AWS, Azure, and Google Cloud all appeared in recorded support videos during the keynote, which suggests the ecosystem either believes the market is large enough to absorb a new competitor, or sees strategic value in validating the CPU demand thesis even as a new rival enters. Probably both.
The competitive landscape reinforces the bet. Nvidia launched the standalone Vera CPU at GTC this month, positioned for agentic orchestration. AMD's EPYC Venice brings 256 Zen 6 cores on a two-nanometer process with a claimed 70 percent generational performance improvement. Intel's Clearwater Forest packs 288 cores on its 18A process. Qualcomm is re-entering the server market. The CPU market has not been this contested in two decades.
What does not hold up as cleanly is the infrastructure prophecy. The argument that agentic AI will drive a fourfold increase in CPU demand per gigawatt is coherent, and Intel and AMD have both signaled CPU price increases of up to 15 percent starting this month — which suggests supply constraints are real. But the agentic adoption curves that would validate Arm's $15 billion projection are assumptions, not data. Enterprises are only beginning to deploy agentic workflows at scale, and the CPU-to-GPU ratio changes that Arm is betting on have not yet been measured in production environments.
The market is pricing Arm's ambition accordingly. $6 billion of incremental revenue — not $15 billion. That gap is not a failure of imagination. It is a rational discount for execution risk on a product that has not shipped yet, in a market that has not yet materialized, against competitors who have been building server CPUs for decades.
Arm is making a real bet. The question is whether the market is right to be this skeptical.
