Nvidia CEO Jensen Huang to CEOs: There is something wrong if your highly-paid engineer does not…

Jensen Huang has a new litmus test for top engineering talent — and it is not the complexity of their commits. Speaking on the All-In Podcast this week, the Nvidia CEO said he would be "deeply alarmed" if a highly-paid engineer spent too little on AI tokens.

The number he has in mind is specific: at least half their annual salary. "If that $500,000 engineer did not consume at least $250,000 worth of tokens, I am going to be deeply alarmed," Huang said. "If that person said $5,000, I will go ape something else." Asked whether Nvidia is trying to spend $2 billion on tokens across its engineering org, his answer was short: "We're trying to."

AI tokens are effectively the new billable hour for software work

Tokens are the unit AI systems use to process text—roughly one word fragment, or about four characters. Every time an engineer prompts Claude Code or OpenAI's Codex to read a codebase, suggest a fix, or generate software from scratch, the model burns through tokens on both ends: input and output.

Running autonomous agents around the clock can push a single engineer past hundreds of millions of tokens a week. At OpenAI's pricing — $15 per million for its most capable model—those numbers translate into real money, fast.At Nvidia's GTC conference in San Jose earlier this week, Huang argued that token budgets should be a formal line item in engineering compensation, not an afterthought. "Every engineer that has access to tokens will be more productive," he said.

He compared an engineer who ignores AI to a chip designer choosing paper and pencil—technically functional, completely indefensible.

Tokenmaxxing: the trend where spending more AI compute signals career ambition

Huang is not the only one drawing this connection. A New York Times report this week found that employees at Meta and OpenAI now compete on internal leaderboards tracking token consumption, with managers factoring AI usage into performance reviews. One OpenAI engineer processed 210 billion tokens in a week alone—equivalent, the Times noted, to reading Wikipedia 33 times over.

The behaviour has acquired a name: tokenmaxxing.The hiring market is already catching up. Thibault Sottiaux, an engineering lead on OpenAI's Codex team, wrote recently that job candidates are asking about token access during interviews—ranking it alongside salary, bonus, and equity as a deciding factor.

Why Nvidia is pushing this idea harder than anyone else

Nvidia's enthusiasm here is not purely philosophical. Its chips are what produce tokens at scale, and enterprise token consumption is, ultimately, GPU demand by another name. But Huang's underlying point stands independent of that: a senior engineer with a large token budget and a fleet of AI agents running in parallel is not one person anymore, in any meaningful sense. Companies that underfund that capability, he implied, are not saving money.

They are just falling behind.