Arm’s stock sinks even as it reveals strong interest in its CPUs for AI servers

Shares of Arm Holdings plc surprisingly lost ground in extended trading today after the chip design firm reported fourth-quarter financial results that surpassed Wall Street’s expectations.

Arm also revealed an encouraging figure relating to its plans to start manufacturing complete central processing units itself. It reported earnings before certain costs such as stock compensation of 60 cents per share, edging past the Street’s target of 58 cents. Revenue for the period came to $1.49 billion, up 20% from a year earlier and above the consensus estimate of $1.47 billion.

Arm said its licensing segment did especially well, with revenue reaching $819 million at the end of the March quarter, up 29% and topping the $781 million analyst consensus estimate. But though sales from the royalty business rose 11% to $671 million, they fell short of the Street’s target of $690 million.

For the current quarter, Arm said it’s looking for sales of about $1.26 billion at the midpoint of its guidance range, just ahead of the Street’s $1.25 billion forecast. It’s also looking for earnings of 40 cents per share, versus the 37-cent-per-share analyst consensus.

Wall Street reacted negatively to the report. Arm’s stock slid more than 6% in after-hours trading, eroding some of the gains made earlier in the day. During the regular trading session, its stock had risen 13%. However, Arm is still one of the best-performing stocks this year, up more than 114% in the year to date.

Arm is best known for designing CPU cores, the parts of a processor that carry out calculations, and various other components, which are sold to other chipmakers to aid in their chip designs. However, the company last year revealed that it’s planning to manufacture its own complete CPU for the first time, and in March it revealed more about that upcoming processor.

It’s called the Arm AGI CPU, and it’s said to provide more than twice as much performance per server rack as Intel Corp.’s rival x86 silicon. Additionally, Arm claims, the AGI CPU can help data center operators reduce hardware costs. The company expects the chip to provide savings of up to $10 billion per gigawatt of data center capacity.

In a conference call with analysts, Chief Executive Rene Haas (pictured) said the company has already generated more than $2 billion in customer demand for its first CPU, stretching through fiscal 2028. It’s set to ensure that Arm’s data center business will soon become its largest segment overall. Haas told analysts that he expects the company to have the largest market share by CPU type by the end of the decade. “We think it’s a market that we can play into in a very large way,” he insisted.

The Arm AGI CPU was co-developed with Meta Platforms Inc. and designed specifically for agentic artificial intelligence systems that can perform business tasks autonomously, with minimal human input. “Data centers are expected to require more than four times the current CPU capacity per gigawatt as agentic AI scales, creating a market opportunity of more than $100 billion by 2030,” Haas added.

Traditionally, Arm has been seen as a leader in the smartphone industry, licensing its instruction-set architecture to companies such as Apple Inc. and Qualcomm Inc. that design chips for low-powered devices. It’s often said that Arm’s chip designs power 99% of the world’s smartphones.

But in recent years, the company has massively expanded its footprint in data center servers, which used to be dominated by Intel’s and Advanced Micro Devices Inc.’s x86 CPUs. Chips such as Nvidia Corp.’s Grace CPUs, Amazon Web Services Inc.’s Graviton CPUs and Google Cloud’s Axion CPUs are all based on Arm designs, which focus on energy efficiency.

The rise of agentic AI has increased demand for CPUs. While most AI training workloads are run on Nvidia’s graphics processing units, they can be overkill for a lot of inference jobs, when fully trained models are run in production.

Tim Arcuri, an analyst with UBS, told MarketWatch that he expects the total addressable market for data center CPUs to reach around $170 billion by 2030. That would “translate to across-the-board upside” for Arm, Intel and AMD, he added. One reason for that is the growing “attach rate” of CPUs to GPUs in AI servers. Traditionally, AI training workloads have required one CPU for eight GPUs, but the ratio for inference is increasingly one CPU for four GPUs.

In addition, Arcuri said, he’s seeing more demand for “standalone CPU-only server racks.”

Arm, Intel and AMD will all benefit from this growing demand for CPUs for AI workloads, but Arcuri said he believes Arm’s instruction set will likely capture the biggest market share. That’s because of the way it’s optimized to scale up core count and throughput while boosting energy efficiency, he explained.

Photos: Arm/YouTube