Story Stocks^®

Arm Holdings (ARM -3%) heads lower despite the company telling Reuters today that its data center CPU market share will swell from roughly 15% in 2024 to 50% by the end of this year. The report comes at a time when The Information noted today that major cloud providers, i.e., Amazon AWS (AMZN), Microsoft Azure (MSFT), and Google Cloud (GOOG), are beginning to show restraint in AI-related spending. The report comes just a few months after these hyperscalers announced or reaffirmed plans to hike their AI spending in 2025, with the total figure expected to reach as high as $320 bln, up an estimated 40% yr/yr.

However, the landscape is changing. The Information stated that some of these Big Tech companies that were quick to pour funds into AI deployments two years ago are starting to take a step back and assess their returns on investment before shelling out additional capital. Meanwhile, lingering fallout from the DeepSeek announcement in January remains. The Chinese startup shook the AI industry when it performed similarly to flagship U.S.-based AI models despite spending only a few million dollars on training. DeepSeek is open-source, like Meta AI, which can undercut the pricing on models from OpenAI and other closed-source alternatives. Enterprises are starting to look at these more seriously as a way to reduce overall AI costs.

Still, just because enterprises are potentially looking into spending less on AI does not mean that they are not steadfast in their commitment to ensuring their respective leadership positions in the industry. In fact, tilting toward cost efficiency could be a net positive for ARM.

• What separates ARM's CPU architecture from x86 alternatives developed by Advanced Micro (AMD) and Intel (INTC) is its efficiency. AI is incredibly power-hungry, which puts pressure on the industry to develop products that can do more but use less power. In February, ARM touched on the DeepSeek breakthrough, noting that it will be a net benefit for AI and ARM primarily because it drives efficiency, making it easier to run AI applications where power is constrained.
• ARM does not make chips but licenses its architecture to chip designers like NVIDIA (NVDA) and Apple (AAPL), then takes a royalty on unit sales. As such, it does not have to ramp production to meet its ambitious data center market share goal. Instead, it only needs its customers to look toward designing more power-efficient using its architecture.
• However, ARM has been looking into creating its own chips. Last month, FT.com reported that ARM was looking into launching chips this year after securing Meta Platforms (META) as one of its first customers. ARM's plans include competing with Qualcomm (QCOM), one of its largest customers, to sell data center CPUs to META. Barring any setbacks, by leaning on licensing while selling its own chips to META, ARM has the potential to upend the AI space and ultimately reach its 50% market share target.

An uncertain economic environment may be dampening the mood around hyperscalers' appetites for AI spending. However, ARM is in an attractive position, touting considerable efficiency advantages over alternative CPU architecture to support a potential run toward cementing its 50% market share in the data center CPU space.