Record revenues, falling shares: what the numbers show

The Cambridge-born semiconductor designer reported fourth-quarter revenues of $1.49bn (£1.09bn) and full-year revenues of $4.92bn for FY2026, according to the company's earnings release on Wednesday. The results marked Arm's third consecutive year of revenue growth above 20 per cent since its return to public markets via a Nasdaq listing in September 2023.

Datacentre revenue more than doubled year-on-year. Licensing income rose. Forward demand for Arm's new AI processors has already surpassed $2bn across the next two fiscal years, the company told investors.

None of that prevented the share price from sliding. Executives conceded during the earnings call that Arm had not secured enough foundry capacity to meet demand for its new CPU products, processors purpose-built for AI datacentres and agentic AI workloads.

"It wasn't the numbers that spooked investors, but concerns about sourcing manufacturing capacity for its next-gen AI processors," said Darren Nathan, head of equity research at Hargreaves Lansdown. "That can be read as yet another signal of white-hot demand for computing power."

Adjusted profit margins remained at roughly 40 per cent once IPO-era stock compensation distortions are stripped out, according to the company's filings, a figure that underscores the underlying financial strength of the business even as operational risks mount.

From blueprints to silicon: the risks of Arm's strategic pivot

For more than three decades, Arm operated as a licensing powerhouse. The company designed processor architectures and collected royalties from the likes of Apple, Nvidia, Qualcomm and Samsung, while never fabricating a chip itself. The model was asset-light, high-margin, and largely insulated from the capital-intensive realities of semiconductor manufacturing.

That model is now changing. Chief executive Rene Haas used the latest earnings to position Arm as a direct participant in the AI infrastructure race. Earlier this year, the company launched its first production CPU in partnership with Meta, a chip designed specifically for data centres running agentic AI systems, where autonomous software tools execute tasks without human involvement.

Arm says its architecture now accounts for roughly half of CPU market share among leading hyperscalers. Google recently announced it would replace processors with Arm-based Axion chips in future TPU systems, as reported by City AM. Nvidia's latest CPU platform is also built on Arm architecture. Amazon Web Services said its Arm-based chip business is generating over $20bn annually.

The commercial opportunity is substantial. But selling production silicon, rather than licensing intellectual property, leaves Arm far more exposed to manufacturing constraints. The company depends principally on TSMC, the Taiwanese foundry that also serves Nvidia, Apple, AMD, and a growing queue of hyperscalers racing to scale AI infrastructure simultaneously.

For a business that built its reputation on capital efficiency, the pivot introduces a category of risk that Arm's balance sheet and operational playbook were never designed to absorb. Memory shortages and foundry bottlenecks are already constraining how quickly the company can fulfil orders, executives acknowledged.

A cautionary tale for UK tech

Arm's experience carries wider relevance for British technology companies weighing similar transitions. The AI boom is creating powerful incentives for IP-focused firms to move closer to physical production, capturing more value per unit sold. Yet the episode illustrates how quickly asset-light advantages can erode when a company enters a supply chain it does not control. Founders and boards considering such pivots would do well to study Arm's capacity constraints closely.

Supply-chain squeeze and what it means for AI infrastructure buyers

The bottleneck at Arm is not an isolated event. Global foundry capacity, concentrated overwhelmingly at TSMC's facilities in Taiwan, is under acute pressure. Amazon, Google, Microsoft, and Nvidia are all scaling AI infrastructure at pace, competing for the same advanced manufacturing slots.

For operators building on or procuring AI infrastructure, the implications are direct. Lead times for next-generation compute hardware are likely to lengthen. Costs may rise. Firms that assumed AI processing capacity would scale linearly with demand now face the prospect of allocation constraints and price premiums.

Memory shortages compound the problem. Arm warned that rising memory costs had weighed on royalty revenues, reflecting tightness across the broader semiconductor supply chain. High-bandwidth memory, essential for AI training and inference workloads, remains in short supply globally.

The situation creates a strategic dilemma for hyperscalers and enterprise buyers alike. Locking in capacity early offers security but ties up capital. Waiting risks being pushed further back in the queue. Neither option is cost-free.

Outlook: China front-loading, smartphone softness, and margin resilience

Several crosscurrents will shape Arm's trajectory over the coming quarters.

Revenue flowing through Arm China has been surging, according to the company's disclosures. Chinese firms appear to be front-loading licensing agreements amid fears that future export restrictions could limit access to Western semiconductor technology. The trend boosts near-term revenues but introduces uncertainty about sustainability if geopolitical tensions escalate or controls tighten further.

Elsewhere, weakness in lower-end smartphones weighed on royalty revenues. Global handset markets remain soft, and Arm's traditional mobile licensing business, still a significant revenue contributor, faces headwinds that AI-driven datacentre growth cannot entirely offset.

Margin resilience, however, remains a notable feature. With roughly 40 per cent of revenues converting into adjusted profit, Arm retains a financial cushion that many hardware-exposed peers lack. The question is whether that cushion holds as the company takes on more production risk and navigates a supply chain stretched to its limits.

Arm's record revenues confirm its centrality to the AI infrastructure build-out. Its capacity constraints confirm something equally important: demand alone does not guarantee delivery. For a company born as a licensing operation in a Cambridge barn, the distance between designing chips and shipping them has never looked wider.