Arm to ship its first in-house chip in 35 years, built with Meta

In its 35 years of existence, Arm has licensed its architectures to thousands of companies, including Apple, Qualcomm, NVIDIA, Samsung and Amazon. All of them built chips on Arm designs, while Arm itself stayed in the background as a supplier of intellectual property. That is now changing: for the first time in its history, Arm is releasing its own processor, and Meta is the first buyer.

This is not just corporate news. It is a shift in a business model that took three and a half decades to form. Arm, which SoftBank acquired for $32 billion in 2016 and which went public in 2023 at a valuation above $60 billion, has always earned its money from royalties and architecture licensing.

Now the company wants a direct share of the silicon market, and it is entering it in partnership with one of the world's largest consumers of computing power. Details about the specific chip are scarce so far. It is unknown which process node is used or who manufactures the physical wafers.

Arm traditionally works with TSMC, which remains the most likely option. Development was carried out jointly with Meta, a company that spends tens of billions of dollars a year on infrastructure for training and deploying its AI models: Llama, advertising algorithms, and recommendation systems for Instagram and Facebook. Why now?

Arm is leaving its traditional niche just as demand for specialized AI chips has hit record levels. NVIDIA dominates the training of large models, but companies are increasingly looking for alternatives for inference: cheaper, more energy-efficient solutions that can scale in data centers without multibillion-dollar electricity bills. Arm architecture has historically led in energy efficiency, which makes it attractive in precisely this segment.

For Arm itself, the move carries obvious strategic risks. The company risks putting itself in competition with its own licensees: Qualcomm, Amazon (Graviton), Microsoft and Google, which have spent years building their own Arm chips. None of these partners will be pleased to learn that their IP supplier is now entering the market for finished products.

On the other hand, it is a logical step for a company that has long been stuck in the royalty trap. Even as the market for Arm devices grows, revenue growth is tightly capped at a percentage of the chip's price. Selling silicon directly opens up a completely different scale of margins.

The partnership with Meta works in both directions. Meta is building its own AI infrastructure and wants to reduce its dependence on NVIDIA. Arm gets an anchor customer and a real-world case study for negotiations with other hyperscalers: Amazon Web Services, Microsoft Azure and Google Cloud.

All of them are looking for alternatives to the H100 and B200 for inference workloads, and an Arm chip with competitive performance and better power consumption could potentially fill that niche. The shift from an IP company to a chip maker is one of the riskiest strategic maneuvers in the semiconductor industry. Historically, few companies have made that transition successfully.

But if Arm pulls it off, that same risk could turn it from a niche royalty supplier into a key player in the next decade's AI infrastructure.

*Meta has been designated an extremist organization and is banned in Russia.