Arm Holdings plc (Arm (ARM)) shares took a hit on Thursday, giving up early gains after the company revealed a frustrating reality: demand for its new AI chips is through the roof, but supply can't keep up. Even after topping fiscal fourth-quarter earnings expectations, the stock slid as executives warned that supply constraints could limit near-term growth, alongside some caution about the smartphone market.

The company said demand for its AI data center CPUs remains red-hot, but it has secured enough supply to meet only about half of current demand. That's a classic good-news-bad-news situation: the product is clearly a hit, but you can't sell what you can't build.

Still, the overall picture Arm painted was one of massive opportunity. Executives highlighted accelerating AI-driven demand, expanding partnerships, and growing adoption of Arm-based infrastructure as the company positions itself at the center of the next wave of cloud and agentic AI computing.

Agentic AI: The CPU's Big Moment

CEO Rene Haas made a compelling case that the shift from human-based AI queries to continuous agent-driven workloads is fundamentally changing the role of CPUs in AI infrastructure. Think of it this way: when you ask ChatGPT a question, a GPU does the heavy lifting. But when AI agents start running autonomously—coordinating tasks, managing memory, enforcing security, and orchestrating accelerators—the CPU becomes the quarterback.

Haas argued that this creates a data center CPU opportunity exceeding $100 billion by 2030. Arm's AGI CPU, launched at the Arm Everywhere event, is designed specifically for this market. Haas said it delivers more than 2x performance per rack compared to x86 platforms and has the potential to cut AI data center capital expenditure by up to $10 billion per gigawatt. That's the kind of efficiency that gets CFOs excited.

The demand is already there. Haas said Arm now sees more than $2 billion in customer demand across fiscal 2027 and 2028 for the Arm AGI CPU—more than double the figure disclosed at launch. The company is scrambling to secure additional wafer, packaging, memory, and testing capacity to support that demand. CFO Jason Child said Arm still expects initial production revenue in the fourth quarter of the fiscal year, while maintaining its long-term outlook of $15 billion in AGI CPU revenue and $10 billion in IP revenue by fiscal 2031.

Haas also stressed that AI infrastructure increasingly requires higher CPU core counts. The AGI CPU already packs 136 cores, and Haas predicted future designs could scale to 256 or even 512 cores as AI orchestration workloads expand. That's a lot of cores, but when you're coordinating armies of AI agents, you need the firepower.

Partnerships Everywhere

Arm's ecosystem story is getting stronger by the quarter. Haas pointed to Meta Platforms Inc. (Meta (META)) as Arm's lead partner and co-developer on a multi-generation roadmap supporting AI workloads for billions of users. That's a big deal—Meta isn't just a customer; it's helping design the future of Arm's AI chips.

Haas also highlighted deployments and partnerships involving SAP SE (SAP (SAP)), Cloudflare, Inc. (Cloudflare (NET)), F5, Inc. (F5 (FFIV)), and SK Telecom. The list of companies building on Arm is getting longer and more diverse.

On the hyperscaler front, Haas said NVIDIA Corp. (NVIDIA (NVDA)), Amazon.com Inc. (Amazon (AMZN)), and Alphabet Inc. (Google (GOOGL)) already use Arm CPUs as head nodes for accelerator systems. Meanwhile, companies including Cerebras, OpenAI, Rebellions, and Positron are adopting the Arm AGI CPU. Google paired its TPU 8t and TPU 8i systems with custom Arm Axion CPUs, and AWS continues scaling Graviton alongside Trainium and Nitro. Microsoft Corp. (Microsoft (MSFT)) is also advancing its Arm-based Cobalt platform for Azure workloads.

The financial impact is already visible. Child said Arm-based server chips deployed by hyperscalers continue driving rapid royalty growth, with data center royalty revenue more than doubling year over year. And Arm maintains nearly 100% market share in data center networking chips like DPUs and SmartNICs. That's a dominant position that's hard to challenge.

Platform Play and Long-Term Bets

Arm's strategy is evolving beyond just licensing IP. Haas said the company now combines traditional IP licensing, compute subsystems, and silicon products into a single compute platform and software ecosystem. Silicon is a new growth vector that complements, rather than replaces, Arm's existing IP business. More than 50 companies support Arm's expansion into silicon, including ecosystem partners across manufacturing, EDA tools, software, and cloud infrastructure.

Child said Arm signed multiple strategic agreements during the quarter, including a long-term AI technology partnership with the Indonesian government and two next-generation CSS licenses covering smartphone and data center networking chips. Annualized contract value grew 22% year over year, reflecting sustained licensing momentum.

Looking ahead, Arm executives expressed confidence in continued growth driven by demand for AI infrastructure. Haas said Arm-based CPUs are becoming central to cloud performance, energy efficiency, and AI economics, and he predicted Arm could become the largest CPU architecture in the data center market by the end of the decade. Child added that the company expects around 20% royalty and licensing growth in fiscal 2027, supported by continued hyperscaler adoption and AI investment.

Price Action

Arm shares were down 8.62% at $216.85 at the time of publication on Thursday, according to market data. The market's message seems to be: great story, but show me the supply.