SAN FRANCISCO — Arm Holdings crossed a $300 billion market valuation as CFO Jason Child laid out the company's plan to move beyond its traditional intellectual property licensing model and into selling finished AI chips. The stock's American Depositary Receipts were up nearly 2 percent Monday morning, extending a year-to-date gain of 152 percent—well ahead of the iShares Semiconductor ETF (SOXX), which has risen 127 percent over the same period.
The licensing business made Arm one of the most capital-efficient companies in semiconductors. Every smartphone on the planet runs on Arm-designed CPU architecture, and the company collects royalties without owning a single fab. The shift to finished chips changes that math entirely.
"Delivering silicon is definitely more complicated," Child said, comparing the new chip business to the existing licensing operation. The remark understates the operational gap. A licensing deal requires lawyers and engineers. A finished chip requires wafer allocations at Taiwan Semiconductor Manufacturing Co. memory supply agreements, power contracts and a distribution stack Arm has never built.
Arm announced its formal entry into AI chips in March with the AGI CPU, a processor designed for AI data centers and agentic AI workloads. The opportunity has expanded faster than the company anticipated. "The opportunity has gotten much larger than what we'd even planned when we started it," Child said, before noting that supply constraints are now more visible too.
The company is projecting more than $2 billion in customer demand for its AI chips across fiscal years 2027 and 2028. That figure is a demand signal, not a revenue guarantee—Child said Arm is starting from scratch in manufacturing and needs a couple of years to build its share of fabrication and memory capacity. "When you're trying to build chips, you have to kind of get in line," he said.
Three specific bottlenecks are slowing the ramp. Memory is the most acute, Child said. TSMC has also discussed wafer and throughput constraints directly with Arm's team. Power availability at AI data centers—where electricity demand per rack keeps climbing—rounds out the list. All three are industry-wide problems, but they hit new entrants harder than incumbents who already hold long-term capacity agreements.
To close those gaps faster, Arm is using acquisitions. The company acquired networking startup DreamBig, a deal that gives Arm proprietary networking intellectual property to pair with its Neoverse compute cores. The strategic logic resembles Nvidia's 2019 acquisition of Mellanox Technologies, which gave Nvidia the InfiniBand networking stack it now uses to sell complete data center systems rather than individual GPUs. Arm has completed roughly 20 acquisitions over the past 20 years, most of them small private companies.
Child said the company evaluates acquisitions on two criteria: whether a target expands Arm's total addressable market, or whether it adds capabilities the company would otherwise have to build internally. He did not rule out larger transactions. "Large M&A is certainly a capability," he said, adding: "Certainly we will evaluate everything."
The DreamBig acquisition fits the first criterion. Arm's Neoverse cores already sit inside servers alongside Nvidia GPUs for AI model training. Adding networking silicon lets Arm offer a more complete compute-communication package—and charge for it as a system rather than a licensed design. That is a structurally higher-margin sale if Arm can execute it, though Child's own comment about the complexity of silicon delivery is the honest caveat.
Arm's existing chip business already benefits from the AI infrastructure build. Its CPUs run alongside Nvidia GPUs in AI training servers, generating royalties on every unit shipped. The new AGI CPU targets a different layer of the stack—AI inference at data center scale for agentic workloads, where the demand on compute is, in Child's words, "for the most part insatiable."
SoftBank owns roughly 87 percent of Arm and is the primary beneficiary of the valuation expansion. The $300 billion market cap makes Arm one of the most valuable semiconductor companies in the world, giving it a currency for acquisitions that most chip designers cannot match. That equity firepower is the practical reason the pivot from licensing to finished chips is now financially credible, even if the fabrication queue problem remains real.
The core risk in the model shift is execution timeline. Arm's licensing business generates revenue the moment a chip partner signs an agreement. The finished-chip business requires years of capacity ramp before it produces meaningful revenue, with capital requirements and supply chain dependencies the company has not managed at scale before. Child's two-year estimate to build fabrication and memory share sets the earliest realistic window for the AGI CPU to contribute materially to Arm's income statement.
