This week, a topic that has been boomeranging around Silicon Valley is back in the spotlight: AI tokens as compensation. The idea is very simple. Companies don’t just give engineers salaries, stocks, and bonuses, they also give them budgets for AI tokens, the computational units that power tools like Claude, ChatGPT, and Gemini. You spend these running agents, automating tasks, and running code. The pitch is that access to more computing makes engineers more productive, and more productive engineers become more valuable. The idea is that it’s an investment in the person who has it.
Nvidia’s leather-jacketed CEO Jensen Huang seemed to capture everyone’s imagination at the company’s annual GTC event earlier this week when he floated the idea that engineers should once again receive about half their base pay in tokens. By his calculations, his top people could be burning through $250,000 a year on AI computing. He called it a recruiting tool and predicted it would become standard across Silicon Valley.
It’s not entirely clear where this idea first came up. Tomasz Tunguz, a well-known Bay Area venture capitalist who runs Theory Ventures and focuses on AI, data, and SaaS startups, whose writing on all things data has earned him a loyal following over the years, spoke about this in mid-February, writing that tech startups were already adding inference costs as a “fourth component of engineering compensation.” Using data from compensation tracking site Levels.fyi, he estimated the top quartile of software engineer salaries to be $375,000. Adding $100,000 to the token will give you $475,000 fully loaded. That means approximately one-fifth of your dollar will be spent on computing.
That’s no coincidence. Agentic AI is rapidly gaining popularity, and the release of OpenClaw in late January significantly accelerated this discussion. OpenClaw is an open-source AI assistant designed to run continuously while you sleep, churning out tasks, spawning subagents, and processing to-do lists. This is part of a broader shift towards “agent” AI, systems that not only respond to prompts but also autonomously perform a series of actions over time.
The practical result is that token consumption has exploded. While someone writing an essay might spend 10,000 tokens in an afternoon, an engineer running a swarm of agents can burn through millions of tokens in a day. It runs automatically in the background without you having to type a word.
By the end of this week, the New York Times had put together a smart investigation into the so-called token maxing trend, finding that engineers at companies like Meta and OpenAI are competing on internal leaderboards that track token consumption. The newspaper reported that generous token budgets are quietly becoming a standard job perk, much like dental insurance and free lunches once were. One Ericsson engineer in Stockholm told the Times that his employer probably spends more on Mr. Claude than on his salary, although it is covered by his employer.
Perhaps tokens really become the fourth pillar of engineering rewards. But engineers may want to hold the line before accepting this as a direct victory. More tokens may mean more power in the short term, but given how quickly things are evolving, it doesn’t necessarily mean increased job security. First, allocating large amounts of tokens comes with high expectations. When a company is effectively funding a second engineer’s worth of computing on your behalf, the implicit pressure is to produce twice as fast (or more).
tech crunch event
San Francisco, California
|
October 13-15, 2026
And there’s an even more troubling problem underlying it. At the point where a company’s token spend per employee approaches or exceeds that employee’s salary, the financial logic of employee count starts to look different to finance teams. When computing is doing the work, it becomes difficult to avoid the question of how many humans need to coordinate.
Jamal Glenn, an East Coast-based Stanford University MBA and former venture capitalist turned financial services CFO, similarly points out that what looks like a perk to a company can be a clever way to increase the apparent value of a compensation package without increasing cash or capital. The token budget is not fixed. I don’t appreciate that. Unlike base salary or stock grants, it does not factor into subsequent offer negotiations. If companies succeed in normalizing tokens as paychecks, it may be easier to keep cash compensation flat while pointing to increases in computing allowances as evidence of investment in employees.
It’s a good deal for the company. Whether it’s a good deal for engineers depends on questions that most engineers don’t yet have enough information to answer.
