Arm Holdings shares have more than doubled this year, driven by robust demand for its CPU designs and solid sales of its new AI chip products. The SoftBank-backed company, now valued at $300 billion, has seen its stock surge bolster its acquisition firepower. Historically, Arm has pursued mostly small-scale acquisitions, such as last year's $265 million purchase of networking startup DreamBig. In a recent interview, Chief Financial Officer Jason Child indicated that while the company will likely continue this acquisition strategy, he did not rule out the possibility of larger deals.

This comes at a time when Arm is no longer merely licensing chip designs to companies like Nvidia and Apple but has begun selling its own self-developed chips, forcing the company to navigate fresh growth opportunities and challenges tied to business expansion. "Delivering physical chips is far more complex than licensing IP designs," Child said. Arm announced the launch of its AI-focused CPU in March, with Meta and OpenAI already signed on as customers. The company reported $4.9 billion in revenue last fiscal year and projects customer demand for this new chip to exceed $2 billion across fiscal 2027 and 2028.

Many players have entered the AI chip development space, including Microsoft, Meta, OpenAI, and Anthropic, alongside Arm. Child stated that the company's timing for launching its dedicated AI CPU is fitting. The viral popularity of OpenClaw, an open-source software used to build AI agents, has sparked a surge in personal AI assistant development. Such applications require cloud service providers to deploy Arm-designed CPUs at scale, utilizing software tools to execute long-running tasks. These workloads are heavily dependent on memory and CPU system scheduling. In servers where developers train AI models, Arm-designed CPUs typically work alongside Nvidia GPUs.

While the surge in CPU demand benefits chipmakers like Intel, AMD, and Nvidia, it also intensifies supply tightness for underlying components such as memory. As a newcomer to chip sales, Arm can only secure limited production capacity from foundries like TSMC. Regarding its self-developed AI chip business, Child noted, "The scale of the market opportunity has far exceeded our expectations when we started this venture. But now we are directly facing the challenges posed by various capacity constraints." Child, who previously worked at Amazon before joining Arm in 2022, also addressed the semiconductor supply chain shortage, whether Arm might follow Intel's lead in issuing new shares, and the relationship between parent company SoftBank and Arm. The following interview has been edited for length and clarity.

Question: Arm's traditional model is asset-light with high profit margins, but now you are selling finished chips. What new financial risks do you foresee?

To produce chips, you have to queue up for capacity. Foundries generally allocate capacity based on a customer's previous year's order volume, adding increments according to their own expansion plans. If you are starting from scratch, your initial capacity allocation will be very small. Ramping up capacity takes years, and we need several years to secure sufficient manufacturing capacity, memory resources, and so on. That is a major challenge. Another challenge comes from rapidly rising prices: demand far exceeds supply. The reality is that without massive financial reserves, it is very difficult to compete in this space. Those making big investments in the industry right now are mostly large cloud providers. They rely on strong free cash flow to support their investments, and their business models are very robust. This makes it hard for smaller companies to enter the market, which is why Groq was eventually acquired by Nvidia. To scale up chip shipments, companies must raise substantial capital and secure borrowing capacity to fund necessary capital expenditures. That is why our financial guidance is conservative, allowing several years for the business to grow gradually.

Question: How do you view the CPU shortage issue?

Computing demand continues to climb. The core bottleneck largely comes down to how much electricity the parties investing heavily in capital expenditure can secure. Chips are only one part of the equation; the entire system is indispensable: you need data centers to house the equipment and power to generate AI output. In my view, demand for compute is almost insatiable. Companies that want to buy chips also need to address tight memory supply; to ensure chip manufacturing, they must secure wafer capacity from TSMC; and finally, they must secure power supply. Arm is just one link in the chain, but from what we observe, there are no signs of computing demand slowing down.

Question: What is the most critical supply chain bottleneck right now, and how can Arm respond?

Memory is the primary bottleneck. At the same time, TSMC has repeatedly mentioned wafer capacity and output capabilities, but there are only a limited number of fabs that can handle advanced process nodes. Many large model companies and cloud providers plan to build data centers with computing power ranging from 1 to 5 gigawatts, consuming enormous amounts of electricity. This raises questions about where to site data centers and whether power can be supplied. The industry generally believes that power supply will remain tight for years to come.

Question: Arm is shifting from pure IP licensing to deeper involvement in chip development. How are your investment priorities changing?

When planning our business, we first evaluate whether we can build capabilities in-house, then determine which areas require external acquisitions. Mergers and acquisitions fall into two main categories: one is to expand the total addressable market (TAM), and the other is to fill gaps in our own development capabilities to help improve profit margins. No company in the semiconductor industry can achieve a fully integrated in-house stack. Nvidia is very close, and Apple is nearly there as well. Broadcom, Nvidia, and Apple maintain high profit margins precisely because they control most of the technology stack. Arm's transformation is still in its early stages, and most past acquisitions have been aimed at expanding market opportunities. The most recent acquisition, DreamBig, is a networking technology company. Arm itself lacks deep expertise in networking, so this deal directly expands our addressable market. Past acquisitions have focused on small and mid-sized private companies. Over the past 15 to 20 years, Arm has completed about 20 acquisitions, basically following this strategy with good results overall, and has rarely pursued large-scale public company takeovers. Of course, we will continue to monitor market opportunities.

Question: Reports say Arm made a last-minute bid for Cerebras last year and also considered a major merger with Marvell. What will be the key considerations for future acquisitions?

We evaluate all opportunities; some targets approach us, some we discover on our own, and we look at all potential targets in the market. Large-scale mergers are viable. Arm is 87% owned by SoftBank, with Masayoshi Son serving as chairman. Looking at SoftBank's past investments, it acquired Graphcore. At that time, a number of companies in the market were developing specialized processors (XPUs), including Graphcore, SambaNova, Cerebras, and Groq, and SoftBank evaluated all of them. Arm and SoftBank have a close relationship, and Arm's CEO Rene Haas also serves as CEO of SoftBank International. How do you coordinate Arm with companies in the SoftBank ecosystem to achieve business synergies? Rene's appointment as SoftBank International CEO makes perfect sense. All assets under SoftBank International serve Son's grand vision: achieving artificial superintelligence (ASI). Rene oversees business operations, mainly focused on chip and system development, addressing needs related to Son's "Stargate" project. Son sets the top-level vision and directs capital allocation and group strategy; Rene is responsible for execution. This collaboration model is highly efficient. When coordinating with SoftBank ecosystem projects, Rene can communicate directly with Son, bypassing multiple layers of approval. The AI industry is iterating at an unprecedented pace, creating enormous opportunities, but it also puts pressure on the entire supply chain, which is why a special operating structure is needed.

Question: Even profitable cloud providers issue new shares when stock prices are high. Could chip companies, or even Arm itself, do the same in the future?

There are no announcements at this time. Generally, when the market experiences a significant pullback, companies tend to buy back shares; when stock prices are at historical highs, they are more likely to issue new shares. Issuing new shares is indeed an option we would consider. Through our relationship with SoftBank, we have access to mature experience in capital operations. For many high-risk, capital-intensive projects, we choose to cooperate with SoftBank, providing IP licensing and design services to complete project development, avoiding taking on heavy debt. SoftBank is also an important customer for our AI-related products, and we will disclose relevant plans later. Through such collaborations, Arm does not need to issue large amounts of debt, which would be very difficult without a partner like SoftBank. Chip companies have built various financing partnerships with Wall Street, and innovative financing structures continue to emerge. What are your observations on this?

The scale of capital being deployed right now is unprecedented. I worked at Amazon from 1999 to 2001 and witnessed the frenzy of the internet era. At that time, the e-commerce outlook was bright, but the user experience was poor; you would place an order and wait one to two weeks for delivery. Many companies secured funding but did not fully utilize hardware capacity, and it took 15 to 20 years to digest the hardware and capital expenditures built during that period. This AI cycle is different: capital investment is immediately deployed into production. The real variable is that business models continue to evolve, and the return on investment for various business segments has yet to be validated. Capital is mainly flowing to companies with strong cash flow, including Amazon, Meta, Google, and Microsoft. Even in an extreme downturn, the worst-case scenario would be writing off large asset impairments, while the core entities would continue to operate normally.