On March 25, Arm unveiled the AGI CPU — its first in-house AI chip. This is a company that has spent 35 years designing chip architectures for others to build. Apple, Qualcomm, Samsung, Google, Nvidia — all license Arm's designs. Arm's business model was neutrality itself: design the blueprints, let everyone build, collect royalties from all sides. Today it announced it would compete with its own customers. And Nvidia, which dominates tech headlines on any given day, was absent from the news entirely.
The £24 Billion Bet
SoftBank acquired Arm in 2016 for £24.3B — the largest Asian takeover of a UK firm. The investment thesis was straightforward: as computing spread to billions of devices, the company that designed the architectures would collect royalties on all of them. Arm licensed designs, partners built chips, SoftBank collected.
But by 2019, Arm was already looking for more. Leaked documents revealed in January 2025 show the company had been working on a strategy to hike prices — a sign that licensing alone wasn't delivering the returns SoftBank needed. By 2020, SoftBank tried a different approach entirely: sell Arm to Nvidia for $40 billion.
The Two-Year No
What followed was one of the most consequential regulatory actions in semiconductor history. Google, Microsoft, and Qualcomm asked regulators to block the deal. Qualcomm told regulators worldwide that the acquisition would harm competition. The UK launched an investigation. The EU opened its own. The FTC sued to block it.
The argument was consistent across jurisdictions: Arm's neutrality was essential infrastructure. If Nvidia owned Arm, it could favor its own designs, throttle competitors' access, or extract unfair licensing terms. The entire semiconductor supply chain depended on Arm being Switzerland — designing for everyone, competing with no one.
In February 2022, Nvidia and SoftBank called it off.
The message from regulators was unambiguous: Arm must remain neutral. The ecosystem depends on it.
The Quiet Build
Then Arm began doing exactly what regulators feared Nvidia would do — but from the inside, where no regulator could intervene.
-
APR 2023Sources: Arm plans to work with manufacturing partners to build a test chip — its first step beyond pure design.
- Aug 2023 SoftBank consolidates ownership. Apple, Samsung, Nvidia, and Intel invest in Arm's IPO.
-
MAY 2024Sources: Arm sets up an AI chip division, aiming to build a prototype within two years.
- Jan 2025 Leaked docs reveal the 2019 price-hike strategy.
-
FEB 2025Two stories break the same week. Sources: Arm plans to launch its own chip. A filing: Nvidia cut its Arm stake by 43.8%.
- Feb 2025 Arm recruits executives from its own customers. SoftBank enters talks to acquire Ampere Computing.
- Jul 2025 Arm CEO Rene Haas says the company is accelerating R&D spending.
Through all of this, Arm kept partnering publicly. In June 2025, Haas sided with Nvidia CEO Jensen Huang on export controls — allies on policy. In October, sources said Arm wanted OpenAI to pair an Arm CPU with Nvidia's and AMD's AI chips — positioned as complementary, not competitive. In November, Arm and Nvidia announced NVLink Fusion, integrating Arm CPUs directly with Nvidia GPUs.
Partnership, partnership, partnership. Until today.
The Numbers
Arm CEO Rene Haas projects $25 billion in revenue by 2031. The breakdown reveals the strategy: $15 billion of that target — more than three times Arm's current total revenue — is expected to come from sales of the AGI CPU. Licensing gets Arm to $10 billion. Competition gets it to $25 billion.
Neutrality was a $4 billion business. Abandoning it might be a $15 billion one.
The first customers confirm the scale of ambition. Not startups. Not second-tier cloud providers. Meta and OpenAI — Nvidia's two most important AI customers.
The Silence
Nvidia has appeared in 83 trending tech stories over the past 30 days — an average of 2.77 per day. Today — the day Arm announced a direct competitor to Nvidia's data center business, with Nvidia's biggest customers as first buyers — Nvidia was absent from the news entirely.
Nvidia likely saw this coming. When it cut its Arm stake by 43.8% in February 2025 — the same month Arm's chip plans leaked — it was selling a position in a company that was about to become a rival. Then it kept partnering anyway, signing NVLink Fusion nine months later.
The public posture was collaboration. The portfolio move was exit.
What the Regulators Built
The deepest irony is structural. Regulators spent two years building a case that Arm's neutrality was essential infrastructure. The FTC, the EU, the UK's CMA all agreed: Arm must remain neutral because the chip ecosystem depends on it. They spent enormous institutional resources to block one company from compromising that neutrality through acquisition.
They were right about the principle. It just didn't matter.
The regulatory system was built to protect Arm's neutrality from external threats — from being acquired and redirected. It has no mechanism for when Arm abandons neutrality voluntarily. No antitrust authority can block a company from competing with its own customers. The two-year regulatory campaign preserved Arm's independence specifically so Arm could, four years later, use that independence to do exactly what the regulators feared.
Not because the regulators were wrong. Because the economics changed. When AI chips became a market measured in hundreds of billions, the value of neutrality — steady licensing royalties — became a rounding error compared to the value of competition.
For 35 years, Arm drew the blueprints and let others build. The regulators who blocked Nvidia's acquisition understood something important: the architect's neutrality was a public good. What they couldn't anticipate was that the architect would look at the buildings going up around it — $500 billion AI companies running on its designs — and decide the blueprints were worth less than the buildings themselves.