Your Signing Bonus Is Now Measured in AI Tokens

Jensen Huang stood on stage at Nvidia's GTC conference this year and said something that should unsettle anyone who still thinks of a salary as a number with a dollar sign in front of it. Companies, he argued, should give their best engineers a token budget worth roughly half their base pay. His math: a top engineer running AI tools at full tilt might burn through $250,000 in compute a year. That's a specific number from a man who sells the compute, and it landed anyway.

A few people are already treating it as gospel. Theory Ventures general partner Tomasz Tunguz has started calling AI inference the "fourth component" of engineering pay, right alongside salary, bonus, and equity. Using Levels.fyi data putting a top-quartile engineer's total comp around $375,000, he estimates a $100,000 annual token budget would shift roughly a fifth of a fully loaded package into pure compute. "Will you be paid in tokens? In 2026, you likely will start to be," he wrote, and the trend backs him up: engineers at Meta and OpenAI reportedly compete on internal leaderboards tracking who burns the most tokens, job listings now advertise generous AI budgets the way they used to advertise unlimited PTO, and one engineer told TechCrunch he spends more on Claude every month than he earns in salary, on the company's dime.

Here's where it gets uncomfortable, and it isn't the price tag. Tokens don't vest. They don't compound. They don't show up in your next salary negotiation as part of your comp history. A $100,000 token budget has a book value of $100,000 this year and exactly zero the day you walk out the door. Put $100,000 in base salary next to it instead, and that number rolls into your baseline forever, feeds your retirement contributions, and anchors every equity grant tied to it going forward. These are not the same instrument wearing different clothes. One builds wealth. The other rents productivity and calls it compensation.

There's a name for what this looks like from the finance side, and it isn't flattering. A package advertised as $375,000 in salary plus a $100,000 token allowance reads, on a recruiting slide, like a $475,000 offer. It behaves nothing like one. Former VC turned CFO Jamaal Glenn has pointed out that token compensation is a clean way to inflate a headline offer number without moving a single dollar of cash or a single share of equity. Which raises the obvious follow-up nobody in the recruiting deck wants asked out loud: if a company is willing to spend a quarter million dollars a year making one engineer twice as productive, how many engineers does it actually still need? That question already has a rough answer taking shape elsewhere in the industry, in the layoff numbers piling up alongside every AI productivity claim.

The comparisons being drawn elsewhere only sharpen the picture. A UK law firm, Shoosmiths, built a bonus pool tied directly to Microsoft Copilot usage: hit a million interactions company-wide, and roughly thirteen hundred employees split the pot. That's not a compensation philosophy. That's a loyalty program for software the firm already pays for, dressed up as generosity. Nvidia's framing is a recruiting pitch. Theory Ventures' framing is an investment thesis. The law firm's framing is a usage incentive. None of these are the same claim, and none of them are obligated to be true at the same time, which is exactly how you end up with three different industries agreeing tokens are the future of pay for three unrelated reasons.

It's worth noticing who benefits from all of them agreeing at once. Nvidia sells the chips. AI labs sell the subscriptions. Everyone loudest about treating compute as compensation has a direct financial interest in normalizing higher per-employee spending on it. That doesn't make the argument wrong. It makes it worth reading the way you'd read a car salesman's opinion on whether you need a bigger engine.

None of which means the access itself is worthless. An engineer with unlimited model access genuinely ships more, faster, and companies willing to fund that access without a leash are signaling something real about how they've adapted. But value and ownership are different questions, and 2026 is quietly answering the second one for an entire industry before most of the people affected have noticed there was a question. A salary is something you keep. A token budget is something you use, on your employer's terms, until the day it isn't there anymore. Ask which one you're actually being offered before you sign for either.