Tech companies are putting AI tokens in your paycheck

Silicon Valley compensation packages used to be simple: base salary, equity, and a year-end bonus. Lately, a fourth line item is quietly slipping onto offer letters: AI tokens.

Instead of just handing over cash and stock, companies are now arming engineers with hefty AI computing budgets to burn through models like Claude, ChatGPT, and Gemini. On March 22, TechCrunch reported that Silicon Valley tech companies are increasingly bundling six-figure token allowances into pay packages, with some totaling $100,000 or more.

From a $100,000 perk to half your salary

Nvidia CEO Jensen Huang pushed the idea straight into prime time at the recent GTC conference. He suggested engineers should receive roughly half their base salary again in AI compute tokens.

By Huang's math, top performers at Nvidia might easily chew through $250,000 a year in model compute. He pitched it as both an essential productivity lever and a powerful recruiting magnet, predicting it would soon become standard across the industry.

He wasn't the first to notice. In mid-February, Theory Ventures founder Tomasz Tunguz pointed out that startups were already treating inference costs as the fourth pillar of engineering comp. Drawing on data from Levels.fyi, Tunguz noted that an upper-quartile software engineer earns around $375,000. Tack on $100,000 in tokens, and the fully loaded package reaches $475,000, meaning one out of every five compensation dollars is literally raw compute.

According to The New York Times, an internal culture of "tokenmaxxing" is already flourishing at companies like Meta and OpenAI, where developers even flex on leaderboards tracking compute consumption. One Ericsson engineer in Stockholm told the paper his monthly Claude tab exceeds his actual salary, paid entirely by the company. Much like free artisan lunches and dental care once were, compute is becoming the modern tech perk.

The catch behind the compute

The explosion in token usage comes down to autonomous agentic workflows. When the open source assistant OpenClaw dropped in late January, running tasks around the clock while users slept, token consumption went vertical. A human drafting a memo burns 10,000 tokens in an afternoon; a fleet of background agents burns millions before breakfast.

Yet accepting tokens as compensation is far from an unalloyed win for workers. Jamaal Glenn, a former VC and now financial services CFO, cautioned that this trend is a clever way for tech firms to inflate the perceived value of an offer without committing real cash or stock.

Your compute allowance doesn't vest. It doesn't appreciate in value. You cannot take unspent tokens to your next employer and leverage them for a bigger salary. Companies can conveniently freeze base salaries while pointing to swelling token allocations as proof they are investing heavily in talent.

Then comes the silent productivity tax. If leadership gives you an extra engineer's worth of compute, the unspoken expectation is that you will deliver double the output. And once your compute bill surpasses your base salary, finance executives inevitably start asking an uncomfortable question: if the algorithms are doing the bulk of the heavy lifting, how many humans are actually required to oversee them?

Why it matters

Turning AI compute into a compensation pillar looks generous on the surface, but we think it cuts both ways for engineers.

For builders, having unrestricted access to the best frontier models is undeniable leverage right now. It lets you prototype faster and automate routine grunt work. But you should not mistake a company compute line item for personal wealth. If your primary skill is simply spinning up token-hungry workflows, the tool doing the work might ultimately diminish your bargaining power rather than protect your role.

For companies, the move is undeniably brilliant. It rebrands standard operating infrastructure costs as flashy compensation perks, inflating top-line offers without triggering long-term cash liabilities or diluting equity pools. It also keeps engineering teams deeply locked into the proprietary AI ecosystems of major cloud and model providers.

We expect token budgets to become commonplace across tech hiring in the coming months. But once the novelty fades, CFOs will start demanding hard returns on all that burned compute. At that point, the engineers who thrive won't be the ones burning the most tokens, but the ones proving that human judgment still matters.