Job candidates are negotiating AI token budgets alongside salaries now. And at least one of them made it a condition of joining.Speaking at GeekWire’s Agents of Transformation event in Seattle on Tuesday, Microsoft EVP Charles Lamanna said a prospective hire told their would-be employer they’d only come aboard if the team was guaranteed a certain dollar amount in AI tokens. Lamanna didn’t name the figure, but put a ceiling on what “generous” looks like: $100 to several hundred dollars of token cost per employee, per day.It’s a striking data point, but not a surprising one. Nvidia CEO Jensen Huang had already floated the same idea at the GTC conference last week — and then sharpened it on the All-In Podcast. Huang said he’d give engineers token budgets worth roughly half their base salary on top of regular pay, and that he’d be “deeply alarmed” if a $500,000 engineer spent only $5,000 worth of tokens in a year. “If that person said $5,000, I will go ape something else,” Huang said.
What AI tokens actually are, and why they’re getting expensive
Tokens are the atomic unit of AI computation — the chunks into which models break down text before processing it. Every prompt entered into Claude or ChatGPT, every function an agentic tool writes overnight, burns through tokens. AI companies charge per million tokens used, and the bills compound fast when agents run unsupervised for hours. The New York Times reported last week that a single Anthropic user ran up a $150,000 Claude Code tab in one month. One OpenAI engineer processed 210 billion tokens — enough text to fill Wikipedia 33 times — in a single week.Lamanna’s argument is purely economic. A fully-loaded engineer costs $500,000 a year. Give them $100,000 in tokens and they become three times as productive — that’s a straightforward return. Withholding that access, he said, is like stripping an engineer of their mouse, email, and Teams on their first day. Technically employed. Functionally hobbled.
Token budgets are quietly becoming the fourth pillar of tech compensation
Venture capitalist Tomasz Tunguz described inference costs last month as a potential fourth component of compensation, sitting alongside salary, bonus, and equity. OpenAI’s Codex engineering lead has noted that candidates are already asking about compute access during interviews. The NYT reported that employees at Meta and OpenAI are competing on internal leaderboards tracking token consumption — a phenomenon now being called “tokenmaxxing.”Lamanna sees the shift moving beyond engineering into financial planning, operations, and other information-heavy roles. The logic is the same across all of them: more tokens, more output. At this rate, leaving a token budget off a job offer won’t be progressive. It’ll just be a red flag.
