SIGNALprophet format
Compute rationing has climbed one more step up the supply chain: it now lives inside the company that makes the chips. The world's most valuable chipmaker, Nvidia — whose graphics processing units (GPUs) are the workhorse of modern AI — cannot fully supply its own engineers. A small group of executives, with Jensen Huang at the center, decides each week which project gets compute, which waits, and which dies.
Nvidia's head of automotive, Xinzhou Wu, said on The Verge's Decoder podcast that the process is recurring and personal: peer teams fight over allocations, and when they cannot agree, Huang arbitrates. Wu also flagged that semiconductor fab capacity — the contracted manufacturing slots at plants like TSMC that turn chip designs into silicon — is its own internal constraint, on top of GPU fights.
The mechanism is portable. Scarcity was first felt by AI customers like OpenAI, Microsoft, and xAI; it has reached the supplier's own org chart. Whoever inside Nvidia wins the weekly meeting wins the next training run; whoever loses waits a quarter.
What gets prioritized? Wu pointed to Huang's "zero trillion dollar business" — Nvidia's shorthand for new markets it hopes will one day be worth trillions — and named autonomous driving (AV) as the lead bet, even though the AV business is still a sliver of data-center revenue. That is capital allocation against a future, not a current result.
When rationing migrates inside the supplier, the binding constraint is no longer the chip. It is one CEO's calendar.
Reported by Sky for Type0, from Jensen Huang still steps in to settle internal fights over Nvidia's scarce AI chips, executive says. Read the original: businessinsider.com