Jensen huang will fire you for not spending enough on ai tokens
Jensen Huang has a new performance metric for Nvidia's top engineers, and it has nothing to do with lines of code or chip designs shipped. It's about how much they're spending on AI tokens. Specifically: if a $500,000-a-year engineer isn't burning through at least $250,000 in tokens annually, Huang says he'd be "deeply alarmed."
The math huang is doing in his head
Speaking on the All-In Podcast Thursday, Nvidia's CEO laid out a blunt equation. Half your salary, minimum, should be going back into AI compute. "If that engineer said $5,000, I would go crazy about something else," Huang said, making clear this isn't a suggestion. It's an expectation baked into how he evaluates talent.
The analogy he reached for was deliberately unflattering: a chip designer who insists on working with paper and pencil. The implication being that under-utilizing AI tokens isn't just inefficient — it's professionally indefensible at Nvidia's level.
When pressed on whether Nvidia itself is spending $2 billion in tokens for its engineering teams, Huang didn't flinch. "We're trying," he said.
Tokens as the fourth leg of the compensation stool
This isn't Huang going rogue. Earlier this week at the GPU Technology Conference, he floated tokens as a recruiting tool, describing a future where engineers receive roughly half their base salary again — in tokens — as a productivity multiplier. "Now it's one of the recruiting tools in Silicon Valley: how many tokens come with my job?" he said.
The broader market is moving the same direction. Tomasz Tunguz of Theory Ventures has described tokens as a potential "fourth component" of compensation, sitting alongside salary, bonus, and equity. Peter Gostev, AI capability lead at Arena, went further, suggesting that OpenAI and Anthropic should build dedicated recruiting pages that list token budgets alongside salary ranges. Thibault Sottiaux, an engineering lead on OpenAI's Codex team, confirmed on X that candidates are already asking about compute access before they ask about stock options.
Sam altman's longer game
The idea that AI compute could function as a form of currency isn't new, even if Huang is the one currently making it operational. Sam Altman floated a version of this in a May 2024 All-In appearance, musing about "Universal Basic Compute" — a world where everyone gets a slice of GPT-7's processing power to use, sell, or donate to cancer research. "What you get are not dollars but this kind of portion, you own part of the productivity," he said.
That was a thought experiment. What Huang is describing is a hiring policy. The distance between those two positions tells you everything about where the industry's head is right now: tokens aren't a philosophical curiosity anymore, they're a line item on the offer letter. And if you're not spending them fast enough, Huang is watching.