This week, a significant topic in Silicon Valley has gained attention: compensating engineers with AI tokens. The concept is simple — along with salary, equity, and bonuses, companies would provide engineers with a budget of AI tokens. These tokens are the computational units that power AI tools like Claude, ChatGPT, and Gemini, allowing engineers to run agents, automate tasks, and process code. The proposition is that more computational power enhances productivity, making engineers more valuable. It’s seen as an investment in the individual holding the tokens.
Jensen Huang, CEO of Nvidia known for his signature leather jacket, captivated audiences at the company’s annual GTC event by suggesting that engineers should receive an amount equivalent to half their base salary in tokens. According to his calculations, top engineers might use up to $250,000 annually on AI compute. He positioned this as a recruiting tool and predicted it would become a norm in Silicon Valley.
The origins of this idea are somewhat unclear. Tomasz Tunguz, a well-known VC in the Bay Area who leads Theory Ventures and focuses on AI, data, and SaaS startups, discussed this in mid-February. He noted that tech startups were already incorporating inference costs as a “fourth component” in engineering compensation. According to data from Levels.fyi, a top-quartile software engineer salary stands at $375,000. Adding $100,000 in tokens brings the total to $475,000 — indicating that about one-fifth of compensation is now in compute.
This shift is no coincidence. Agentic AI has been rapidly advancing, with the release of OpenClaw in late January spurring the discussion. OpenClaw is an open-source AI assistant designed to operate continuously, handling tasks and managing to-do lists while users are offline. This is part of a broader movement towards “agentic” AI, where systems independently execute sequences of actions over time.
As a result, token consumption has surged. While writing an essay might require 10,000 tokens in an afternoon, engineers managing a network of agents can expend millions daily, all without manual input.
Recently, the New York Times examined the trend of “tokenmaxxing,” noting that engineers at companies like Meta and OpenAI are competing on internal leaderboards tracking token usage. According to the report, generous token budgets are quietly becoming a typical job perk, similar to dental insurance or free lunches in the past. An Ericsson engineer in Stockholm mentioned he likely spends more on Claude than his salary, but his employer covers the cost.
While tokens might become the fourth pillar of engineering compensation, engineers should be cautious before accepting it as an outright benefit. More tokens could mean increased short-term power, but as things rapidly change, it doesn’t assure job security. A large token budget comes with high expectations, as companies effectively fund the equivalent of a second engineer’s compute, expecting double the output (or more).
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
Moreover, there’s a deeper issue: if a company’s token expenditure per employee matches or surpasses their salary, the finance team may reevaluate headcount logic. As compute handles more work, the necessity for human coordination becomes questionable.
Jamaal Glenn, who has transitioned from a VC to a financial services CFO with a Stanford MBA, also acknowledges that what seems like a benefit might actually be a tactic for companies to enhance the perceived value of a compensation package without increasing cash or equity — elements that appreciate over time. Unlike base salary or equity grants, a token budget doesn’t vest, doesn’t grow, and isn’t considered in future salary negotiations. If tokens become normalized as pay, companies might maintain flat cash compensation while citing a larger compute allowance as a sign of investment in their workforce.
This arrangement benefits companies, but whether it serves engineers well depends on factors most engineers currently lack the information to fully evaluate.
