Micron Rally Reflects AI Demand Outrunning Semiconductor Supply

Daniel Pilling framed the rally in Micron and other semiconductor names as a supply problem created by unusually fast AI adoption meeting a market with long production lead times and concentrated suppliers. Asked by Ed Ludlow whether the basic picture is that Nvidia is sold out and demand is outpacing supply for data centers “on Earth” as well as more speculative space-based data-center ideas, Pilling agreed.

His case rests on three linked claims. First, AI demand is “viral,” with growth he compared to Zoom during Covid and described as “unparalleled,” pointing to Anthropic’s numbers as evidence. Second, adoption remains very low relative to the potential market. Pilling said Sands Capital estimates that only about 2% or 3% of the world’s roughly one billion office workers are using AI tools today. Third, that demand is arriving in a semiconductor market where suppliers have long lead times, oligopolistic structures, and pricing power.

That combination, in Pilling’s view, makes a prolonged supply constraint likely. The visible market context was Micron: Bloomberg’s on-screen headline said Micron had received a street-high target as UBS saw a “$1.8T giant,” while an intraday panel showed Micron up 16.78%.

Caroline Hyde pressed Pilling on whether investors should broaden their exposure beyond Nvidia and the hyperscalers that own more of the vertical stack, including Alphabet, Microsoft, and Amazon. Pilling’s answer was yes: he said Sands Capital owns memory companies such as SK Hynix and Samsung in Korea, and argued that memory will become an increasingly significant bottleneck.

He also pointed to semiconductor capital-equipment companies, naming ASML and Lam Research as businesses that provide equipment to foundries. The investment logic he described extends outside chips as well. Data-center electricity suppliers should be part of the analysis, he said, naming Bloom Energy as one business Sands owns. He also said CPUs are “back,” with Arm particularly interesting to the firm because “agentic AI needs a lot of CPUs.”

The result is a broader stack thesis rather than a single-stock AI thesis. Pilling described “many, many winners” across memory, chip equipment, power, CPUs, and companies operating full AI infrastructure stacks. The common thread is not merely exposure to AI, but exposure to bottlenecks created as AI workloads scale faster than physical infrastructure can be supplied.

Ludlow asked why chip stocks rallied globally after news from Huawei about “logic folding,” described on air as a new method of passing signals through transistors that could move development outside Moore’s Law and reduce dependence on extreme miniaturization. Bloomberg’s lower third summarized the market reaction as “CHIP STOCKS RALLY ON HUAWEI OPTIMISM.”

Pilling said the enthusiasm made particular sense in China and Asia because China lacks access to ASML and to leading-edge chips from Taiwan Semiconductor. Any announcement that suggests China might build its own leading-edge chips is therefore “much appreciated.”

But he was skeptical of the technology as presented. The core idea, he said, has been around for a while: stacking chips on top of one another is something Taiwan Semiconductor has pursued for roughly a decade. The hard part is not the concept but execution. Pilling identified two problems: the chips must actually be stacked, and heat dissipation becomes very difficult to solve.

His view was not that China lacks AI capability. Asked directly about China’s AI prowess and domestic supply chain, Pilling said Sands Capital is “very impressed” by what China has been doing. The constraint, in his view, remains manufacturing. China does not have ASML or lithography manufacturing capabilities that would allow it to build leading-edge chips, which he put at above roughly 5 nanometers. As long as that remains true, he said, China will struggle to compete with Nvidia, SK Hynix, and Micron.

Pilling estimated that China may be a decade away from closing that lithography gap, noting that it took Western nations 15 to 20 years to build something like ASML. The distinction matters: he credited China’s progress in AI, while arguing that the semiconductor supply chain’s hardest manufacturing bottlenecks still limit China’s ability to match the frontier.

Ludlow turned the discussion to Nvidia’s post-earnings math: the company’s Blackwell and Rubin opportunity from calendar 2025 to calendar 2027, hyperscaler capex, and how much of that becomes Nvidia revenue and free cash flow. Pilling said Sands Capital had done the math, but pointed to a longer-dated figure he considered equally important.

He cited Jensen Huang as having discussed $3 trillion to $4 trillion of industry capex by 2030. Pilling called that an “enormously large number,” but said Nvidia’s share across training and inference is believed to be around 60%. If both numbers hold, Pilling said Nvidia could generate close to $40 of earnings per share and nearly $1 trillion of free cash flow by 2030.

On that basis, he said Nvidia would trade at about five times earnings for that period. He acknowledged the assumptions require very large growth from companies such as Anthropic and that the capex number looks “very, very large” today. But he added that Huang has a record, in Pilling’s view, of hitting the numbers he discusses. Sands Capital remains “very positive” on that outlook.