Hyperscaler Backlogs Are Approaching $2 Trillion—and AI Bears Still Need an Answer

$AMZN, $MSFT, $GOOGL and $ORCL now have a combined contracted backlog approaching—and potentially exceeding— $2 trillion . That remains one of the strongest counters to the AI overbuild thesis.
Hyperscalers have incentives to emphasize demand, but Sam Altman’s admission that OpenAI underestimated its compute needs carries different weight: it comes from a major buyer of capacity, not the companies selling it.
Jordi Visser, citing a Dwarkesh Patel essay, argues that scarce compute could allow the labs producing the most output per watt to capture disproportionate value. Under that framework, Anthropic and OpenAI become beneficiaries of the shortage rather than its casualties.
The key test is whether the backlog remains intact through Q3 earnings.