A concept that has been circulating in Silicon Valley gained significant attention this week: using AI tokens as a form of employee compensation. The premise is simple—instead of offering engineers just salary, equity, and bonuses, companies would also provide them with an allocation of AI tokens. These tokens are the computational credits needed to operate systems like Claude, ChatGPT, and Gemini. Engineers could use them to run automated agents, handle tasks, or process code, with the argument that greater access to computing power boosts productivity, making those engineers more valuable. Essentially, it's framed as an investment in the individual's capabilities.

EMBED_PLACEHOLDER_0

Jensen Huang, Nvidia's CEO, notably captured industry interest when he suggested at the company's recent GTC event that engineers might receive an additional amount in tokens roughly equal to half their base salary. He estimated that his top personnel could utilize up to $250,000 worth of AI compute annually, presenting this as a strategic recruiting advantage and forecasting it would become a common practice across the tech sector.

The origins of this idea aren't entirely clear, but Tomasz Tunguz, a well-known Bay Area venture capitalist at Theory Ventures who specializes in AI, data, and SaaS, discussed it in mid-February. He observed that tech startups were beginning to incorporate inference costs as a "fourth component to engineering compensation." Based on data from Levels.fyi, he noted a top-quartile software engineer salary of $375,000. Adding an estimated $100,000 in tokens brings the total compensation to $475,000, meaning approximately one-fifth of the package is now allocated to compute resources. This shift aligns with the rising trend of agentic AI, a movement significantly accelerated by the late-January release of OpenClaw.

OpenClaw is an open-source AI assistant built to operate continuously, managing tasks, creating sub-agents, and working through lists autonomously, even while the user is offline. It represents a move toward "agentic" AI—systems that execute sequences of actions independently rather than merely responding to individual prompts. A direct result has been a massive surge in token usage. While someone drafting an essay might use 10,000 tokens in an afternoon, an engineer operating a network of agents can consume millions in a single day, all automatically and in the background.

By this weekend, coverage highlighted the emerging "tokenmaxxing" trend, reporting that engineers at firms such as Meta and OpenAI are participating in internal leaderboards that track token consumption. According to reports, substantial token budgets are quietly becoming a standard employment benefit, akin to dental insurance or complimentary meals in the past. One engineer at Ericsson in Stockholm mentioned that his Claude usage likely exceeds his salary, though his employer covers the cost.

While tokens may indeed become a standard part of engineering compensation, engineers might consider pausing before viewing this as an unambiguous benefit. Although more tokens can translate to greater short-term capability, the rapid pace of change doesn't automatically guarantee enhanced job security.

For starters, a generous token allocation carries high expectations. If a company is funding computational resources equivalent to a second engineer on your behalf, there's an implicit pressure to deliver at least double the output. Beneath that lies a more complex issue: when a company's token expenditure per employee nears or surpasses their salary, the financial perspective on headcount begins to shift for the finance department. If the compute is performing the work, it raises inevitable questions about how many human coordinators are truly necessary.

Jamaal Glenn, a Stanford MBA and former venture capitalist now serving as a CFO in financial services, adds that what appears to be a perk could be a strategic way for companies to enhance the perceived value of a compensation package without actually increasing cash or equity—the components that genuinely accumulate value for an employee over time. A token budget doesn't vest, appreciate, or carry into future salary negotiations like base pay or equity grants. If tokens become normalized as pay, companies might more easily justify stagnant cash compensation by pointing to a growing compute allowance as evidence of investing in their team.

That scenario represents a favorable arrangement for the company. Whether it benefits the engineer depends on questions most engineers currently lack the information to answer.