Hyperscalers are sitting on a $2.3 trillion time bomb of pricing power (Save this).
Milk Road AI Twitter · Milk Road AI (@MilkRoadAI) · 2026-07-31
Milk Road AI argues that combined cloud backlogs across Microsoft, Amazon, Google, and Oracle reaching $2.3 trillion — up from $800 billion one year ago — represent locked-in pricing power that will drive operating leverage as legacy contracts renew at higher market rates.
Appears in
Extraction
Topics: cloud-backlogshyperscaler-pricinggpu-rental-ratesai-infrastructure-investment
Claims
- Combined cloud backlogs across Microsoft, Amazon, Google, and Oracle grew from roughly $800 billion to over $2.3 trillion in one year.
- H100 one-year GPU rental rates rose 63% from approximately $1.70 to $2.77 per hour despite newer chips already available, an unusual pattern in tech where aging hardware typically gets cheaper.
- Daily token usage in China exceeded 140 trillion calls, over 1,000 times the volume of two years prior.
- Combined 2026 capex guidance across the four hyperscalers is $740–770 billion, nearly double the prior year, and is being deployed against already-contracted demand rather than speculative future demand.
- Microsoft's CFO confirmed the company is capacity constrained, meaning demand exceeds what can currently be delivered.
- As large backlogs renew at higher prices, incremental revenue will drop largely to the bottom line since underlying infrastructure is already built and paid for.
Key quotes
H100 one year rental rates went from about $1.70 an hour in October to roughly $2.77 by now, a 63% jump, even with newer chips already on the market.
That capital isn't chasing hypothetical future demand but rather chasing contracts that are already signed and locked in, which is a very different risk than typical growth spending.
That's a setup where profits could grow faster than spending does, which almost never happens in a capex-heavy business, and it's the kind of operating leverage that historically leads to major re-ratings once investors actually notice it.