Discussions are spreading in Silicon Valley about adding artificial intelligence (AI) tokens to annual compensation packages, alongside salary, stock options and bonuses. Some see it as a new reward tool that could boost engineer productivity. Others raise concerns that companies may be trying to lean on compute resources instead of cash compensation.

On March 22 local time, IT outlet TechCrunch reported that talk has been spreading rapidly in Silicon Valley about paying AI tokens as compensation. The AI tokens refer to compute units or usage budgets used to run generative AI and AI agents such as ChatGPT, Claude and Gemini. The core logic is that engineers can use them to automate repetitive work and generate and modify code, raising productivity and ultimately creating higher value.

The person who set off the debate was Nvidia CEO Jensen Huang (젠슨 황). He recently made remarks at the annual developer event GTC suggesting engineers should receive additional AI tokens worth about half their base pay. Huang said top talent could use $250,000 (about 377 million won) a year in AI compute resources, adding that it could also be seen as a way to improve competitiveness in recruiting.

The spread of what is known as agentic AI underpins the trend. The open-source AI assistant OpenClaw, released in late January, drew attention for continuing tasks while users sleep and creating sub-agents to handle work. As AI that performs tasks autonomously spreads, token consumption is also rising quickly. The explanation is that it is not just tens of thousands of tokens for a single essay, and an engineer running multiple agents at once can burn through millions of tokens a day.

In related coverage, the New York Times reported that at some companies including Meta and OpenAI, engineers are competing over token usage through internal leaderboards. That suggests token budgets are becoming a new workplace benefit. An Ericsson engineer in Stockholm, Sweden, told the New York Times the cost of Claude he uses could be more than his annual pay, but the company is covering it.

Some also argue it is hard to view AI tokens as an immediately favorable form of compensation for engineers. More tokens may allow access to more compute resources and results in the short term, but they could also increase the likelihood of demands for greater productivity. If a company effectively provides one person with compute resources comparable to a "second engineer", a structure could form in which it expects matching speed and output.

A more fundamental issue is that companies' cost calculations could change. If per-employee token spending reaches a level similar to or exceeding annual pay, finance departments will naturally ask how many human workers are needed. That is why concerns are emerging that if compute resources take over a significant part of the work, humans may end up only coordinating them.

Jamal Glendo, a former venture capitalist, also said AI tokens may look like a perk but could in practice be a way to enlarge the appearance of a compensation package without cash or equity. He said token budgets do not vest, do not rise in value over time, and are hard to have recognized in the next salary negotiation after a job change in the way base pay or stock compensation is. If companies succeed in entrenching tokens as part of compensation, they could more easily claim they are "investing more in employees" based on expanding compute budgets even if cash compensation stands still.

Ultimately, AI tokens could be a practical tool to raise engineer productivity, while also carrying the possibility of becoming a new means for companies to reshape compensation structures. It remains unclear whether tokens will take hold as a new standard in annual packages in Silicon Valley, but the debate at least is read as a sign that AI is beginning to change the value of labor and the standards of compensation.