Arm Holdings has spent several years telling investors that its chip designs would move out of phones and into the buildings where artificial intelligence is trained. On November 5 the company produced the revenue to support the claim, and a forecast that did not.
Revenue for the quarter ended September 30, the second quarter of Arm’s fiscal 2027, reached $1.44 billion, up 27% from a year earlier and a record for a second fiscal quarter. Licensing and other revenue was $773 million, up 50%. Royalties, the per-chip payments that arrive long after a design is licensed, were $670 million. Revenue tied to data centers doubled from a year earlier.
Profit was harder to read. GAAP operating income was $201 million. On a non-GAAP basis, which excludes share-based compensation and acquisition costs, operating income was $638 million. Arm also said its partners reported shipping more than 10 billion chips based on its architecture in the period, the first time that figure has crossed that line.
The forecast is where the quarter lost its shine. Arm guided third-quarter revenue to about $1.54 billion, roughly $10 million above the $1.53 billion analysts had expected, and put adjusted earnings per share at 56 cents. For a company whose shares have been priced as a direct beneficiary of AI data center construction, a beat that small reads as an admission that the near-term pipeline is not accelerating. The stock fell 5.28%.
The structure of Arm’s business explains why its numbers move the way they do. Licensing revenue is negotiated deal by deal and recognized when a customer commits to a design, so a handful of large agreements can swing a quarter. Royalties are paid per chip shipped, which means they arrive a year or more after the license is signed and then track the volume of devices sold. Data center processors carry far more value per unit than the microcontrollers and phone chips that built the royalty base, so a shift in mix can lift revenue without a matching rise in unit shipments.
Management put three numbers in front of investors to describe where that shift leads. Arm expects its share of the server market at the largest hyperscale cloud providers to exceed 60% by the end of 2026. It said revenue from a processor line it calls AGI CPU could reach as much as $2 billion across fiscal 2027 and fiscal 2028, with first commercial deliveries in the fourth quarter of fiscal 2027. Both claims rest on designs that customers have licensed but not yet shipped at scale.
That timing matters. A forecast that depends on products entering production in eighteen months asks investors to underwrite a manufacturing ramp, a customer qualification cycle and a competitive response, none of which Arm controls. What the company holds in that negotiation is its instruction set, which software is written against, and the ecosystem of tools built around it. Its exposure is that the largest buyers of server chips have their own design teams and their own reasons to want a second supplier.
Reuters reported that orders for custom processors designed by cloud providers are growing, a trend that helps Arm because those chips are frequently built on its architecture. The same report noted that royalties from handsets remain tied to the replacement cycle, which has been sluggish for several years. Phones still supply the bulk of the royalty base, so even a doubling in data center revenue takes time to change the shape of the total.
The gap between the two profit measures is worth keeping in view. Arm’s adjusted figure is more than three times its GAAP operating income, a spread that reflects how much of its compensation is paid in stock. That is normal for a company of its age and sector, but it means the headline margin depends on which measure an investor chooses, and the choice is not neutral when the shares trade at a multiple that assumes sustained growth.
There is a further wrinkle in how licensing behaves late in a cycle. Customers facing uncertainty about their own volumes sometimes delay new licenses and extend existing ones, which supports revenue in the near term and defers the royalty stream that would have followed. A 50% jump in licensing is a strong number. Whether it represents new design wins or pulled-forward renewals is a question the disclosure does not answer directly.
The stock reaction suggests investors are no longer paying for the story alone. Arm has argued for years that the AI buildout would eventually show up in its royalty line rather than only in its licensing line, and the doubling in data center revenue is the first real evidence. What the December quarter will show is whether that growth is broad or concentrated in a few customers whose spending plans are now the subject of intense scrutiny.
For now the company has a record quarter, a forecast that cleared expectations by the width of a rounding error, and a valuation that requires the second half of its story to arrive on schedule. The question investors will carry into the next report is whether Arm’s share of the server market is being won on merit, or rented from customers who intend to design their own chips once the volumes justify it.
