SAN FRANCISCO — Arm Holdings licenses technology to semiconductor designers that powers virtually every cellphone in the world and many other products. Now, the company has its own chip to sell.
The company, which is a British unit of the Japanese group SoftBank, announced this Tuesday plans for the first silicon product that Arm will design and market since its founding in 1990. It is a microprocessor aimed at data centers that run artificial intelligence tasks.
Meta, owner of Facebook, helped develop the chip and became its first user. Other early customers include AI pioneer OpenAI, Arm said.
Continues after advertising
The new offering marks a drastic change in its business model. The company pioneered the concept of selling intellectual property rather than physical products, charging fees and royalties per chip from hundreds of companies that license Arm’s microprocessor architecture — the equivalent of “blueprints” for designing chips.
Pierre Ferragu, an analyst at New Street Research, called Arm’s shift toward selling chips “the most significant strategic shift in the company’s history.”
Nvidia, which has become the world’s most valuable publicly traded company thanks to its AI chips, has focused attention on the processors it sells to perform specialized tasks with high numerical performance, essential to the development of AI systems.
But microprocessors that handle other functions are gaining importance as technology companies offer so-called AI agents — technologies capable of automating a wide range of tasks. Jensen Huang, CEO of Nvidia, recently predicted that this type of chip could become a multi-billion dollar business for the company.
Arm’s move risks angering some of its biggest licensees, who could become both rivals and customers. In addition to Nvidia, Arm’s customer base includes Qualcomm and major AI players like Amazon, Microsoft and Ampere, whose chips are used by Oracle. All of these companies have been developing chips based on Arm technology.
Mohamed Awad, executive vice president of cloud and AI at Arm, played down the idea that the company will compete directly with its customers. According to him, the new product is aimed at AI computing tasks that only a few cloud giants have the resources to design on their own.
Continues after advertising
“You can’t just buy a product like this on the market today,” Awad said. “Meta and others asked us to go out there and build this.”
But Jim McGregor, an analyst at Tirias Research, said some level of tension with customers is inevitable as Arm comes under pressure to grow revenue beyond what the licensing business can sustain.
“Arm has an aggressive strategy to increase its revenue,” said McGregor. “That doesn’t necessarily make everyone happy.”
Continues after advertising
Arm’s plan to sell its own chips was revealed by Financial Times in February 2025. But the company had already been considering this step for years, according to testimony from Rene Haas, CEO of Arm, in December 2024, in a trial in US federal court arising from a lawsuit filed by Arm against Qualcomm over licensing terms linked to a startup purchased by the San Diego company. In September, a federal judge in Delaware ruled in favor of Qualcomm.
The new product announced on Tuesday, the Arm AGI CPU, has up to 136 computing elements, known as cores (“cores”), offering what the company describes as an extreme increase in the amount of calculations per watt of power consumed. Awad estimated that adopting the technology could save $10 billion in building one of the latest AI data megacenters, which could cost up to $50 billion.
He said that potential customers have already received samples of the chip and should begin using it in computing tasks by the end of the year.
Continues after advertising
c.2026 The New York Times Company
