Nvidia's Arm deal bets that AI computing will run on every device

Nvidia agreed to buy Arm for $40 billion, joining its AI chips with processor designs licensed across the industry. Neutrality and regulators now set the value.
Arm's licensees have shipped 180 billion chips to date, and Nvidia now wants to own the designs inside them. The deal, agreed on 13 September with Arm's Japanese owner, is valued at $40 billion: $21.5 billion in Nvidia stock, $12 billion in cash and up to $5 billion more tied to Arm's performance.
The price buys reach, and the reach comes with a condition. Nvidia built its AI business on data center GPUs, while Arm's designs run in phones, cars, connected devices and a growing number of servers. Owning both would give one company a computing platform for AI from the cloud to the edge and the power to set terms for a layer that much of the device industry licenses. Value in AI hardware concentrates where chips, software and developer ecosystems meet, and this deal shows that regulators, not only markets, will decide how much of it one company may hold.
Neutrality is the asset Nvidia is buying and the one it could destroy. The Cambridge company licenses its processor designs to hundreds of chipmakers around the world, many of them Nvidia's competitors, and its value rests on treating them all alike. Nvidia has promised to keep the open licensing model and customer neutrality, keep the headquarters in Cambridge and build an AI research center there around an Arm-powered supercomputer. Those promises now go before approval in the UK, China, the European Union and the US, a process expected to take about 18 months. The UK review opens in hostile territory, because an Arm co-founder has already called the deal a disaster for Cambridge, the UK and Europe, citing sovereignty and US export controls.
AI-first companies should expect to build on hardware from fewer, larger suppliers. Nvidia plans to add its GPU and AI technology to Arm's licensing catalog, and if it does, machine learning on devices could get standard accelerators and tools, which would lower the cost of shipping models to the edge. Dependence is the price of that convenience. Chip startups that build AI accelerators on Arm cores would license from a company owned by a rival, and software teams tied to one vendor's stack lose bargaining power. Portability across hardware belongs in the first release, and it is a question raised early in any look at the AI-first companies Fanaura backs.
Arm's largest licensees hold the counterargument. If they respond by speeding up work on open alternatives such as RISC-V, the neutral layer Nvidia is paying for starts to lose its pull before the deal even closes. The settling evidence comes after approval: whether Nvidia's GPU designs actually appear in Arm's licensing catalog, the test of whether the deal spreads AI computing or concentrates it.
The bid confirms that AI has become the organizing purpose of the chip industry. Suppliers will consolidate, so our thesis on where AI value settles favours companies that win whichever chip runs underneath. Whoever ends up owning Arm, the safest software is the kind that does not care.
Building an AI-native company? Apply in five minutes. →
More from Fanaura
Building machine learning software that runs well on any chip, from cloud to device? Send us the project.
Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.
Get the news by email
One short note each weekday. No marketing, and you can unsubscribe at any time.