For more than three decades, Arm Holdings has been the quiet architect behind nearly every smartphone processor on the planet. It designed the blueprints. Others built the chips. That division of labor made Arm indispensable — and deliberately non-threatening to the giants who paid for its intellectual property. Now Arm is tearing up the arrangement that made it one of the most important companies in technology.
On Monday, Arm announced it will design and sell its own central processing unit, a complete custom chip rather than the instruction set architectures and core designs it has historically licensed to partners like Qualcomm, Apple, Samsung, and dozens of others. The first customer: Meta Platforms, which will deploy Arm’s custom silicon in its sprawling data centers. The product carries the internal designation of the Arm Total Compute platform, and it represents the most aggressive strategic pivot in the Cambridge-based company’s history, as CNBC first reported.
This isn’t a minor product extension. It’s a full-scale entry into a market Arm deliberately avoided for its entire existence.
The implications ripple in every direction — toward Intel and AMD, which dominate data center processors; toward Qualcomm and MediaTek, which license Arm’s designs for mobile and increasingly for PCs and servers; and toward Nvidia, which uses Arm architecture in its Grace CPU line paired with its AI accelerators. Arm is no longer content to collect royalties from the sidelines. It wants a piece of the silicon itself.
CEO Rene Haas framed the move as a natural evolution during a press briefing Monday morning. “Our partners have done extraordinary work building on our architecture, but we see an opportunity to deliver something purpose-built for the workloads that are defining the next era of computing,” Haas said, according to CNBC. He was careful not to characterize the move as competitive with existing licensees, though that’s precisely what it is.
Meta’s involvement signals that the demand side of this equation is real. The social media and AI conglomerate has been on a furious infrastructure buildout, committing over $60 billion in capital expenditures for 2025 alone, much of it directed at data centers powering its Llama large language models and AI-driven products across Facebook, Instagram, and WhatsApp. Meta has long sought to reduce its dependence on merchant silicon suppliers, experimenting with custom chip efforts internally — its MTIA (Meta Training and Inference Accelerator) program has been running for years — and now apparently willing to let Arm take the lead on the CPU side of its server racks.
Why Meta? And why now?
The answer lies in the economics of hyperscale computing. Companies like Meta, Google, Amazon, and Microsoft collectively spend hundreds of billions annually on infrastructure. Even marginal improvements in performance-per-watt or performance-per-dollar translate into savings measured in billions. Amazon already builds its own Arm-based Graviton processors through its Annapurna Labs subsidiary. Google has its Axion processors. Apple, of course, designs its own Arm-based M-series chips for Macs and its A-series for iPhones. The trend is unmistakable: the largest consumers of compute want custom silicon tailored to their specific workloads, not general-purpose chips designed to serve everyone adequately and no one optimally.
Meta, however, has lacked the deep in-house silicon design capability of an Apple or an Amazon. Its MTIA accelerator chips are narrowly focused on inference tasks. For the general-purpose CPU that orchestrates everything in a server — managing memory, handling networking, running the operating system, coordinating with accelerators — Meta has relied on AMD’s EPYC and, to a lesser extent, Intel’s Xeon processors. Arm’s offer is essentially this: let us build you a custom CPU optimized for your workloads, and you won’t need to hire and retain the thousands of engineers required to do it yourself.
That’s a compelling pitch. It’s also a direct shot across the bow of AMD and Intel.
AMD shares fell 3.2% in midday trading Monday. Intel, already battered by years of strategic missteps and manufacturing delays, dropped 2.7%. Arm’s own stock, listed on the Nasdaq since its September 2023 IPO, surged 8.4% by early afternoon. Investors are reading the tea leaves clearly: if Arm can convert even a fraction of hyperscaler CPU demand into direct sales rather than licensing revenue, the financial upside is enormous. Arm’s current royalty model generates revenue that’s a small percentage of the total chip price. Selling the chip itself — or at least the complete design that a foundry like TSMC manufactures — captures far more value per unit.
The financial math is straightforward. Arm reported $3.9 billion in revenue for fiscal year 2025, a record, but still modest relative to the value of the chips built on its technology. The total market for Arm-based chips exceeded $280 billion in end-product value last year. Arm captures roughly 1-2% of that in royalties. Moving up the value chain, even selectively, could multiply its revenue per chip by an order of magnitude.
But the risks are substantial. Arm’s licensing business depends on trust. Qualcomm, Samsung, MediaTek, and others pay Arm because it’s a neutral supplier of intellectual property, not a competitor. The moment Arm starts selling finished CPU designs — competing directly for the same data center sockets its licensees target — that neutrality evaporates. Qualcomm, which has been locked in a bitter legal dispute with Arm over the terms of its architectural license (a case that went to trial in late 2024 and remains partially unresolved), will view this as confirmation of its worst fears. So will others.
There’s historical precedent for this kind of channel conflict, and it rarely ends cleanly. When Google launched its Pixel phones, it strained relationships with Android OEM partners. When Amazon released its own branded products, third-party sellers on its marketplace cried foul. When Intel briefly tried to compete with its own customers in the finished-system market years ago, it retreated quickly. Arm is betting that the data center opportunity is large enough — and differentiated enough from its mobile licensing core — to manage the tension.
“This is a calculated risk,” said Stacy Rasgon, a semiconductor analyst at Bernstein, in a note to clients Monday. “Arm has clearly decided that the TAM expansion opportunity in data centers outweighs the channel conflict risk in mobile. Whether their licensees agree remains to be seen.”
The technical details of Arm’s CPU offering remain sparse. The company said the chip will be fabricated on TSMC’s advanced N3E (3-nanometer enhanced) process node, suggesting performance and efficiency characteristics competitive with the best current server processors. Arm indicated the design incorporates its latest Neoverse V-series cores, custom-configured for Meta’s workload profile, with particular emphasis on AI inference throughput, memory bandwidth, and energy efficiency. Data center operators are obsessed with power consumption — electricity is often the single largest operating cost for a hyperscale facility — and Arm’s architecture has long held an advantage over x86 in performance per watt.
Meta’s infrastructure chief, Santosh Janardhan, offered a brief statement endorsing the partnership. “We’re always looking for ways to optimize our infrastructure for the workloads that matter most to us,” Janardhan said, per CNBC. “Working directly with Arm on a purpose-built CPU gives us a level of customization that off-the-shelf solutions can’t match.” The phrasing — “off-the-shelf” — is a pointed reference to AMD and Intel’s standard server product lines.
The broader context makes this move even more significant. The data center CPU market is undergoing a structural shift away from Intel’s decades-long dominance. AMD clawed back massive share starting around 2019 with its EPYC line. Amazon’s Graviton processors demonstrated that Arm-based chips could handle enterprise and cloud workloads at scale. Microsoft has developed its own Arm-based Cobalt 100 chip for Azure. Google’s Axion, based on Arm’s Neoverse cores, entered production last year. The x86 duopoly is fracturing, and Arm architecture is the primary beneficiary.
Until now, Arm benefited from this trend indirectly — more Arm-based server chips meant more royalties. But the company clearly concluded that indirect participation wasn’t enough. SoftBank, Arm’s majority shareholder, has been pushing for faster revenue growth to justify the company’s towering valuation, which stood at roughly $150 billion as of Friday’s close. At approximately 38 times forward revenue, Arm is priced for a trajectory that licensing royalties alone may not deliver.
And then there’s the AI factor. Every major technology company is racing to build infrastructure for artificial intelligence workloads. Training large models requires massive GPU clusters, yes, but inference — actually running those models to generate responses, recommend content, or process images — is increasingly CPU-bound for many tasks, or at least requires tight CPU-accelerator coordination. A custom Arm CPU designed specifically for Meta’s AI inference pipeline could offer meaningful advantages over a general-purpose AMD or Intel chip that must serve a thousand different customer profiles.
Nvidia’s position here is interesting. The company’s Grace CPU, paired with its Hopper and Blackwell GPUs in so-called Grace-Hopper and Grace-Blackwell superchips, uses Arm’s Neoverse cores under license. Nvidia and Arm have maintained a close partnership, especially after Nvidia’s failed $40 billion acquisition attempt collapsed in 2022 under regulatory pressure. If Arm starts selling its own CPUs to hyperscalers, it could complicate that relationship too — though Nvidia’s primary value proposition is its GPU technology and CUDA software stack, not the CPU companion.
The server market isn’t the only front. Arm has been expanding aggressively into automotive, IoT, and PC processors. Its architecture powers Qualcomm’s Snapdragon X Elite chips for Windows laptops, a product line that has gained meaningful traction since launching in 2024. If Arm’s direct CPU ambitions extend beyond data centers into PCs or edge computing, the competitive dynamics get even messier.
For now, the company insists the direct CPU effort is focused exclusively on hyperscale data centers. “We are not competing with our partners in mobile, in automotive, or in client computing,” Haas said, according to CNBC. “This is about serving a specific set of customers with unique requirements that are best addressed through a direct engagement model.” The qualifier “for now” hung unspoken in the air.
Wall Street’s initial reaction was enthusiastic but not unanimous. Morgan Stanley analyst Lee Simpson raised his price target on Arm shares, citing the “significant revenue upside” from direct chip sales. But Jefferies analyst Jaime Zabaleta cautioned that “channel conflict is a slow-burning fuse” and that Arm’s licensing revenue could face pressure if partners begin hedging their commitments. RISC-V, the open-source instruction set architecture that has been gaining traction in certain segments, stands to benefit if Arm’s licensees start looking for alternatives to reduce their dependence on a company that’s now also a competitor.
RISC-V remains years away from being a credible threat in high-performance server CPUs. But the mere existence of an alternative gives Arm’s licensees negotiating power they didn’t have before — and motivation to invest in it that they might not have felt before Monday’s announcement.
The production timeline is aggressive. Arm said it expects the Meta-specific CPU to enter volume production in the first half of 2027, with initial engineering samples shipping to Meta by late 2026. That suggests the design is already well advanced, meaning Arm has been working on this in stealth for at least two to three years. The secrecy is notable. In an industry where major product plans typically leak months in advance through supply chain chatter, Arm and Meta managed to keep this under wraps.
One question looms over everything: is this the beginning or the end? Will Arm build custom CPUs for Google, Microsoft, and other hyperscalers too? Or is Meta a singular partnership, a proof of concept that Arm can point to while continuing to collect licensing fees from everyone else?
The answer almost certainly is that Meta is the beginning. The economic logic is too compelling, the hyperscaler demand too strong, and SoftBank’s growth expectations too aggressive for this to be a one-off. Arm has crossed the Rubicon. It’s a chip company now, not just a chip design company. And the semiconductor industry — from Santa Clara to Hsinchu to Cambridge — will be reckoning with the consequences for years to come.
WebProNews is an iEntry Publication