AI hardware continues to see strong demand, driven by the growing need for inference and always-on agents such as Muse, Dots, Autopilot, and Grok. These applications require increasing compute power, leading to higher rental pricing for H200/B200 and spot compute remaining scarce. The shortage of available power is exacerbating the compute shortage, requiring companies to get creative with squeezing optics, bringing more power behind the meter, improving utilization, and hunting for stranded capacity.

While hardware pricing is rising faster than expected, Micron’s FY27 supply/demand remains tight, and 2027 HBM pricing is materially above 2026. Despite these challenges, Goldman’s Rich Privorotsky remains constructive on AI hardware, citing the ability of earnings to keep growing into valuations as long as the “music is playing.”

Privorotsky’s view on hardware is straightforward: as long as demand for AI applications continues to drive growth, companies can overcome supply constraints and maintain their earnings momentum. While concerns about the eventual peak of the cycle exist, Privorotsky believes that valuations have already priced in some of this risk, making for an attractive investment opportunity.