Nvidia's Huang pitches AI tokens on top of salary as agents reshape how humans work - CNBC
Jensen Huang's Next Compensation Package: Tokens for Every Engineer
Jensen Huang has a new idea about how to pay engineers. Not just salary. Not just equity. Tokens.
At Nvidia's GTC conference this week, Huang outlined a vision where every engineer at his company gets an annual token budget — and not a small one. "They're going to make a few hundred thousand a year as their base pay," he said. "I'm going to give them probably half of that on top of it as tokens so that they could be amplified 10 times."
The tokens in question aren't cryptocurrency. They're the basic units of computation that AI models consume — the currency of inference, the fuel that runs AI agents. An engineer with a token budget can run AI models, deploy agents, and automate workflows without hitting the usage limits that currently throttle productivity in token-constrained environments. More tokens means more agent time means more output amplified.
The framing is compensation, but the infrastructure story is about access. Right now, most enterprise AI deployments cap individual usage. Token budgets would give engineers dedicated compute allocations that don't disappear when a quarterly budget runs out. "Every engineer that has access to tokens will be more productive," Huang said. That claim is either obvious or profound depending on how much compute they actually get.
The economic logic is straightforward from Huang's perspective. Nvidia sells the chips. The chips run the models. The models are accessed via tokens. If token budgets become a standard part of engineering compensation, Nvidia sells more chips. Huang is in the rare position of being able to describe his own product as a benefit to the people he wants to recruit.
Whether other companies follow depends on whether token budgets actually produce the productivity gains Huang claims. The "10x amplification" figure hasn't been independently verified. But the idea that compensation will increasingly include compute allocations — not just cash and equity — is a genuine shift in how companies think about talent infrastructure.
Silicon Valley is already asking about it. "It is now one of the recruiting tools in Silicon Valley," Huang said. "How many tokens come along with my job." That's the sentence that matters. The answer to that question, at companies that adopt this model, could become as standard as base salary and signing bonus.
Huang also forecast $1 trillion in computing demand through 2027, saying Nvidia would be short of that figure. The company posted $215.9 billion in fiscal 2026 revenue last month, up 65% year-over-year. Data center revenue specifically reached $62.3 billion, up 75%.
The token compensation story is separate from the product announcements, but it's the one that lands as infrastructure news. Every company building AI agents will eventually face the same question Nvidia is now answering: how do you allocate compute to the people who need to run those agents? Huang's answer is to make tokens a line item on the pay stub.
Sources: