Arm’s AI CPU Orders Double to $2 Billion as Smartphones Weaken

Arm was down about 8% intraday during the interview, even as Bloomberg’s one-year visual showed the stock up roughly 76%. The business tension underneath that move is sharper than a simple demand slowdown: Rene Haas said Arm is seeing weakness in smartphones, but described AI data center demand as strong enough to change the scale of the company’s opportunity.

Haas put the latest quarter at roughly $1.5 billion in revenue, a level he said “not long ago” would have been an annual revenue figure for Arm. A Bloomberg graphic listed fourth-quarter revenue at $1.49 billion, up 20% year over year, with licensing revenue of $819 million, up 29%, and royalty revenue of $671 million, up 11%.

The larger claim was forward-looking. Haas said Arm had visibility into about $1 billion of orders for its Arm AGI CPU in its forecast at the time of the earnings call. Over the following five weeks, that doubled to $2 billion.

That demand sits alongside a longer target. Haas referred to a $15 billion goal by FYE31, which he translated for viewers as calendar 2030, and said Arm is “very confident” it is on track to that number. Relative to the company’s current revenue base, he described the opportunity as transformational.

The smartphone weakness is real, but Haas argued it is not where Arm’s highest-value exposure sits. Much of Arm’s volume, he said, comes from the premium smartphone segment, where royalty rates are “quite rich” because customers are using Arm’s version 9 architecture. The slowdown Arm is seeing is mostly at the lower end of the market, where the royalty contribution is much smaller.

That distinction matters because Arm still reports substantial licensing and royalty revenue, but Haas’s growth story is moving away from the broad handset unit cycle and toward data center compute. He said Arm’s data center business doubled year over year and linked the quarter’s strength to demand for the new AGI CPU.

Rene Haas described a shift in data center workloads around AI agents: more automated queries hitting data centers and requiring fast answers. In his account, those workloads create an operational burden that accelerators alone do not handle — agent management, orchestration, scheduling, and related work.

Haas was explicit that GPUs may accelerate parts of AI computation, but they do not manage the orchestration layer he is describing. “This is not something an accelerated GPU can manage,” he said. The result, in his view, is that demand for CPUs in the data center is “exploding.”

That is the context for Arm’s decision to go beyond supplying intellectual property. For 35 years, Haas said, Arm delivered products through IP — “the blueprint” customers used to build chips based on Arm technology. He pointed to existing Arm-based data center deployments from Amazon, Google, Microsoft, and Nvidia: Amazon’s Graviton, including a recently announced Graviton 5; Google Axion; Microsoft Cobalt; and Nvidia Vera.

Arm is now also selling its own chip. A Bloomberg graphic summarized the move as “Arm to sell its own chips for first time,” adding that the AGI CPU will draw 300 watts of electricity, Meta will be the first major consumer, and Taiwan Semi will produce the chips.

“Demand is certainly not a problem,” Haas said.

Meta is the first major customer highlighted for the AGI CPU, but Rene Haas described demand as broader than one anchor account. He named OpenAI, Cerebras, SK Telecom, Rebellions, SAP, and F5 Networks, and said demand is appearing across network infrastructure and the general push to put more CPUs inside the data center.

Arm is not positioning the AGI CPU as only a standalone chip. Haas described it as part of a system, with partners including Super Micro, Lenovo, and ASRock building the systems and racks customers can order.

The rack-level claim was central to his explanation of demand. Haas said these are 36-kilowatt air-cooled racks, and that customers can get twice the performance in the same power envelope compared with a comparable x86 rack. He connected that to Arm’s long-standing reputation for power efficiency, but presented it as more than a familiar brand attribute: it is one of the factors catalyzing demand.

The constraint, as Caroline Hyde put it, is the one repeatedly heard across AI infrastructure: demand may be strong, but supply determines how quickly companies can deliver. Haas said Arm has supply for the first $1 billion of orders it discussed on the earnings call. The additional $1 billion of demand brought the total to $2 billion, and Haas said Arm is working with the supply chain around that larger number.

He named TSMC and memory suppliers SK Hynix, Micron, and Samsung as companies Arm is working with. The picture he gave was not a demand problem, but a supply-chain coordination task created by order visibility that had doubled in five weeks.

Haas’s broader claim was that the timing risk is different from a short-lived product cycle. “This is not perishable demand,” he said. If a particular window closes, he argued, the need for compute does not disappear. That is the basis for treating the $15 billion FYE31 target as a long-game infrastructure opportunity rather than a narrow launch-window bet.

Rene Haas framed his additional role leading SoftBank International as coordination across assets that could matter to Arm, not as a separate job detached from the chip company. Caroline Hyde pointed to SoftBank’s chip-related portfolio, including Ampere and Graphcore, and asked how SoftBank’s broader role in chips changes Haas’s work.

Haas said there is “a lot of synergy” across SoftBank’s activities. He cited recent SoftBank announcements around a large 10-gigawatt data center facility in Portsmouth, Ohio, involving work with the U.S. Department of Energy and SoftBank Energy. He also named Ampere and Graphcore as portfolio companies whose work could fit with Arm.

The relevance, in Haas’s account, is the overlap among compute, chip design, portfolio companies, energy infrastructure, and data-center projects. He said Masa asked him to help with “orchestration and coordination” across the companies. From one angle, he said, it may look like he has two jobs; from another, the overlap creates coordination benefits on both sides.