Nvidia CEO Jensen Huang used his keynote at the GTC AI Conference on March 20 to propose a compensation model that treats AI compute as a form of pay. Engineers, Huang suggested, should receive AI tokens — units of compute access — on top of their base salaries, with allocations potentially reaching half of an engineer’s annual compensation. For a senior engineer earning $500,000, that would mean an additional $250,000 in AI compute credits annually.
The proposal reflects Huang’s broader thesis that AI agents will fundamentally reshape workforce structure. Nvidia currently employs 42,000 people — what Huang described as “biological employees” — and anticipates supplementing them with “hundreds of thousands of digital employees” in the form of AI agents. In this model, an engineer’s productivity becomes a function not just of skill and time but of how much compute they can deploy. AI tokens become the mechanism for distributing that compute.
The concept has already gained traction in Silicon Valley recruiting. Companies competing for AI talent report that compute access has joined equity, salary, and signing bonuses as a negotiating lever. Goldman Sachs estimates that AI could automate 25 percent of US work hours, a projection that gives Huang’s framing additional weight: if agents handle a quarter of the work, the engineers directing those agents need compute budgets, not just salaries.
Huang’s pitch aligns with Nvidia’s commercial interests. More AI tokens in circulation means more demand for Nvidia’s GPU infrastructure, creating a feedback loop between workforce compensation and hardware sales. The company’s data center revenue has driven its market capitalization past $3 trillion, and embedding compute access into standard compensation packages would expand the addressable market for its chips from corporate IT budgets into payroll systems.
The open question is whether AI tokens represent a genuine shift in how technical work is valued or a rebranding of existing cloud compute credits. Engineers at companies like Google and Meta already receive substantial internal compute allocations. What Huang is proposing — making that allocation explicit, portable, and denominated as compensation — would require new frameworks for taxation, vesting, and valuation that do not yet exist. The IRS has not issued guidance on AI tokens as compensation, and no major employer has announced adoption of the model Huang described.
