Arm Holdings Revenue Climbs 20% as Data Center CPU Demand Surges, Shares Slide 7% After Hours

Arm Holdings reported fiscal fourth-quarter revenue of $1.49 billion on Wednesday, a 20% year-over-year increase that beat analyst expectations, but shares tumbled roughly 7% in after-hours trading as investors weighed supply chain constraints against an otherwise upbeat outlook for the company's data center CPU business.

The quarter ended March 31 came in ahead of the LSEG-compiled analyst consensus estimate of $1.47 billion. Non-GAAP earnings per share rose 9% to 60 cents, topping the 58 cents analysts had forecast.

Despite the earnings beat, the after-hours selloff erased much of the stock's regular-session gains. Arm shares had closed at a record high of $237 on Wednesday, extending year-to-date gains to 117% before the reversal.

For the first quarter of fiscal year 2027, the company guided for earnings of 40 cents per share, plus or minus 4 cents, on revenue of $1.26 billion, plus or minus $50 million — modestly ahead of the FactSet consensus of 37 cents per share on revenue of $1.25 billion. The company also said mobile market unit growth will be flat or slightly negative in fiscal year 2027, a disclosure that weighed on sentiment.

The results arrive as central processing units, long overshadowed by graphics processors in AI infrastructure buildouts, have regained strategic relevance. Intel CEO Lip Bu Tan noted on an April 23 earnings call that the CPU-to-GPU ratio in AI racks was previously 1-to-8, but has shifted to roughly 1-to-4 as agentic AI workloads proliferate — and could reach parity in the future.

Advanced Micro Devices CEO Lisa Su offered a similar assessment on AMD's earnings call Tuesday. Su said she now expects the CPU server total addressable market to grow at a rate exceeding 35% annually, reaching over $120 billion by 2030. "Agents are really driving tremendous demand in the overall AI adoption cycle," Su told CNBC on Wednesday.

Arm CEO Renee Haas underscored his company's positioning within that landscape. "Whether it's Nvidia, whether it's Amazon, whether it's Google, the very largest and most prevalent accelerators by volume are the TPU, it's Trainium, and it's Rubin. ... Those all connect to Arm," Haas said on the earnings call. The company said Arm-based CPUs represent more than 50% share among top hyperscalers.

A newer and closely watched element of the Arm story is its first in-house data center CPU, unveiled in March at the company's Arm Everywhere event and designed specifically for agentic AI workloads. At launch, Arm said it had line of sight to more than $1 billion in customer demand over the next two years.

Management has since doubled that figure. The company now sees over $2 billion in customer demand across fiscal years 2027 and 2028, though executives noted they are maintaining the original $1 billion commitment while supply chain capacity is secured — a caveat that contributed to the post-earnings stock pressure.

Haas framed the in-house chip as a response to customer pull rather than a strategic pivot away from licensing. "The primary reason we did this," Haas said, "was that our customers asked for it. At the end of the day, we are responding to customer demand in a market."

The company also said hyperscalers could potentially reduce AI data center capital expenditures by up to $10 billion per gigawatt through the use of Arm-based CPUs, a pitch that carries weight as the industry scrutinizes infrastructure spending efficiency.

Arm's longer-term revenue target remains $15 billion by the end of fiscal year 2031, and management has indicated the in-house chip business is not expected to cannibalize existing licensing and royalty revenue streams.

How quickly Arm can expand supply chain capacity to match accelerating demand will be the central execution question heading into the back half of 2026.

Disclaimer