The artificial intelligence supercycle is transitioning from a speculative phase driven by chipmakers to an industrial execution phase centered on physical infrastructure. As the market digests massive capital expenditures for large language models (LLMs), demand for foundational infrastructure continues to outpace supply, with the global AI infrastructure sector projected to compound at nearly 25% annually through the end of this decade.
Industry forecasts highlight three primary bottlenecks: thermal management, scalable inference access, and electrical baseload. Vertiv, DigitalOcean, and Hut 8 have emerged as key players addressing these constraints. Vertiv (NYSE: VRT) captures a dominant share in liquid cooling for high-density GPU clusters, boasting 30% year-over-year revenue growth and a profitable, multi-billion-dollar backlog. DigitalOcean (NYSE: DOCN) focuses on a scalable inference cloud for mid-sized enterprises, generating $900 million in revenue last year (up 15% year-over-year), with sales from clients spending over half a million annually surging 76%. Hut 8 (NASDAQ: HUT) manages a 1,000-megawatt production portfolio, enabling rapid deployment of energy to AI projects while bypassing utility grid delays, with 2025 sales hitting $235 million—a 45% year-on-year increase.
Meanwhile, Goldman Sachs analyst Jim Covello recommends investors shift from chipmakers to hyperscalers like Amazon, Microsoft, Alphabet, Meta, and Oracle. He argues that hyperscaler stock multiples have compressed due to investor skepticism about returns on AI spending, while chip stocks (Philadelphia Semiconductor Index up nearly 150% over the past year) appear expensive. Covello outlines two favorable scenarios for hyperscalers: strong returns on AI investments boosting stock prices, or a pullback in spending improving cash flow—both would benefit hyperscalers relative to chipmakers. The main risk is continued heavy spending without clear returns, which would keep hyperscaler stocks under pressure while sustaining chip demand.
