Arm Holdings Launches First In-House Silicon, Targets $15 Billion by 2031

Arm Holdings launched the AGI CPU — the company's first internally designed and manufactured processor — at a San Francisco event on March 25, 2026, sending shares up 16.3% and crystallising a $164,300,000,000 market capitalisation for the British chip-architecture firm. Meta Platforms is the lead customer and co-development partner, with OpenAI, Cerebras, Cloudflare, and SK Telecom among the named launch partners. Commercial systems are already orderable from ASRockRack, Lenovo, and Supermicro.

For 35 years, Arm's business was the licence. It designed instruction-set architectures and physical IP blocks, collected royalties, and left the silicon to partners — Qualcomm, Apple, Samsung, and dozens of others who paid precisely because Arm stayed out of their markets. That model generated roughly $4,000,000,000 in annual revenue in 2025. The AGI CPU abandons the posture entirely. Arm is now a chip company, competing for the same data-centre CPU sockets it once supplied only as blueprints. The closest historical parallel is arguably Intel's mid-1980s exit from DRAM into microprocessors — a moment when an IP-based identity was traded for a product one, with consequences the industry is still living with.

The AGI CPU is built on TSMC's 3nm process and features a dual-chiplet design with 136 Neoverse V3 cores per processor, running at a 300-watt thermal design power. Arm claims the chip delivers 2x the performance-per-rack compared with current x86 servers at a fraction of the power draw — a claim that, if validated in production deployments, would carry significant weight given that power density has become the binding constraint on hyperscale data-centre expansion. (The performance figures are Arm's own; independent third-party benchmarks have not yet been published.) Volume shipments to Meta and OpenAI are expected later in 2026, with Arm projecting fiscal year 2028 as the first year of meaningful AGI CPU revenue. By 2031, the company targets $10,000,000,000 from its traditional IP and CSS licensing business and $15,000,000,000 from AGI CPU sales — a combined $25,000,000,000 in annual revenue, against a stated total addressable market of $100,000,000,000 for agentic-AI CPUs by 2030. Chief executive Rene Haas set those figures publicly at the launch event, alongside earnings-per-share guidance of $9 for 2031.

Rene Haas, CEO at Arm Holdings, told Data Center Knowledge on March 25, 2026: 'It is a massive market. We think there is room for lots of new players. We are not really too worried about that. We just know the market opportunity is very large [and] quite underserved.' Jensen Huang, CEO at Nvidia, said at GTC 2026, as reported by IEEE Spectrum: 'Finally, AI is able to do productive work, and therefore the inflection point of inference has arrived.' Huang's framing matters here — Nvidia sold its remaining equity stake in Arm as recently as February 2026, and the two companies' relationship has shifted from what analysts have described as a partnership to a 'symbiotic rivalry', with Arm's CPU now positioned to coordinate across the same accelerator ecosystem Nvidia dominates.

The immediate pressure falls on Intel and AMD. Arm's performance-per-watt claims target the precise metric on which x86 architecture has been most vulnerable since the AI inference cycle began displacing training as the dominant data-centre workload. Qualcomm, reading the same signals, has accelerated its pivot toward RISC-V, reportedly acquiring Ventana Micro Systems to build a data-centre ecosystem that avoids both the licensing cost and the new competitive friction with Arm. For hyperscalers not yet committed — Amazon and Google among them — the AGI CPU introduces a credible third option alongside x86 incumbency and custom in-house silicon. Arm's decision to make systems immediately orderable through Lenovo, Supermicro, and ASRockRack shortens the evaluation cycle considerably. SoftBank, which retains approximately 90.0% of Arm's shares, holds the majority of the financial exposure to whether the revenue targets materialise — a concentration of ownership that gives the launch an unusual asymmetry between strategic ambition and shareholder accountability.

Watch three signals in the coming months. First, the performance-per-watt data from Meta's initial AGI CPU deployments: if independent benchmarks confirm Arm's 2x rack-efficiency claim at scale, competitive pressure on Intel and AMD could accelerate materially before year-end. Second, counter-announcements from Intel and AMD — Arm's brief puts a 60-to-90-day window on likely responses, and any competitive silicon disclosures in that window will define how much pricing and positioning latitude Arm actually holds. Third, FY28 revenue guidance updates: Arm has named that fiscal year as the inflection point for meaningful AGI CPU contribution, and any revision — upward or downward — to that timeline will function as the clearest early signal of whether the product-company transition is tracking. The $15,000,000,000 target is not modest for a company that earned $4,000,000,000 last year. The architecture of ambition is clear. The execution, as always, is the thing.