From My Perspective, the Market Has Overreacted to Hyperscaler Credit Spread Widening. TL;DR: Spot Pricing for Rented GPU Compute Is Much Higher Than Contract Prices, Meaning Hyperscalers Are Undershooting Earnings While Operating Cash Flow Accelerates, an Underappreciated Funding Source for AI Capex

English summary

The fact that spot GPU rental prices are at least twice contract prices is the missing link in the hyperscaler credit discussion and the only fundamental factor behind this selloff. Several private companies plan to increase compute spending at least 2x per GPU after contracts expire, and some have publicly discussed this. As contracts expire, hyperscaler growth rates will continue to accelerate because the re-pricing of their compute installed base will be higher. Using a mix of estimates and actuals, hyperscaler operating cash flow growth is expected to accelerate from 31% in Q1 2026 to 50% in Q2. This acceleration should continue for the rest of the year, and it is not reflected in estimates—from my perspective, estimates incorrectly model Q3 as a slowdown. Some data: consensus estimates may be for hyperscalers and new cloud to add 25-35 gigawatts in CY28 (using a range, because building data centers is difficult, and many new cloud plus labs are still private). At $60 billion per gigawatt, that would be $1.5-2.2 trillion in capex. ...