$250,000. That's how much AI compute Nvidia wants each of its engineers burning through every year - on top of their regular salary.
During an appearance on the All-In Podcast recorded at GTC 2026 last week, Nvidia CEO Jensen Huang laid out a vision where AI tokens (the basic units of text that AI models process and generate) become a standard part of employee compensation. His math is simple: a $500,000 engineer should consume at least $250,000 worth of AI tokens annually. "If that $500,000 engineer did not consume at least $250,000 worth of tokens, I'm going to be deeply alarmed," Huang said.
The analogy he reached for was blunt. An engineer refusing to use AI would be like "one of our chip designers who says, 'Guess what? I'm just going to use paper and pencil. I don't think I'm going to need any CAD tools.'"
$2 Billion a Year on Tokens
With roughly 42,000 employees, Nvidia's target token spend works out to around $2 billion annually. When asked directly about that number, Huang's response was simply: "We're trying to."
The goal is a 10x productivity multiplier per engineer. Huang described tokens as "now one of the recruiting tools in Silicon Valley" and predicted that companies will soon advertise token budgets alongside salary and equity in job listings.
This isn't theoretical. Nvidia already uses AI extensively in its own chip design process. Reinforcement learning systems generate standard cell libraries overnight. Internal language models trained on decades of Nvidia's design history help newer engineers navigate complex architectural decisions. At GTC 2026, chip design partners Cadence, Synopsys, Siemens, and Dassault Systemes all announced they're building Nvidia-powered AI agents specifically for chip design workflows.
What This Signals for Everyone Else
Huang's framing matters because it puts a dollar figure on something most companies are still figuring out: how much should we actually spend on AI tools per employee?
Most knowledge workers today spend somewhere between $20 and $100 per month on AI subscriptions - ChatGPT Pro at $200/month is considered expensive. Nvidia is talking about $20,000+ per month per engineer. That's a different universe, but the direction is clear. Huang is betting that the return on AI compute will be so high that not spending on it becomes the irrational choice.
The infrastructure numbers back up the ambition. Huang noted that a 1-gigawatt data center using Nvidia's upcoming Vera Rubin chips could generate 700 million tokens per second, a 350x improvement over current hardware. More supply means lower token prices, which means these budgets become practical for a wider range of companies.
For the average AI tool user, this is a useful benchmark. The CEO of the company that builds the hardware running most AI models thinks half of an engineer's compensation should go toward AI compute. Even if your company lands at a fraction of that ratio, the implication is that most organizations are dramatically underinvesting in AI tools for their teams right now.