OpenAI: AI Tokens Revolutionize Tech Compensation

AI as a New Compensation Element

Greg Brockman, president of OpenAI, recently reflected on the tktk days during which CEO Sam Altman was ousted. This introspection comes at a time when the importance of AI computation in tech job compensation is steadily increasing. AI tokens are emerging as a new component of salaries in the tech sector, adding to traditional elements such as salary, bonuses, and stock options. CFOs now face new challenges as AI inference costs become a significant part of tech budgets.

Silicon Valley has always competed to attract talent with increasingly rich compensation packages. Now, a fourth element is making its debut: AI inference. As generative AI tools integrate into software development, the cost of running the underlying models—known as inference—becomes a productivity driver and a budget line item that financial leaders cannot ignore.

The Race for AI Resources

Software engineers and AI researchers within tech companies are already vying for access to GPUs, with this AI computing capacity being carefully allocated based on the most important projects. Now, some job candidates are starting to ask about the AI computing budget they will have access to if they decide to join the company.

Thibault Sottiaux, technical lead of OpenAI's AI coding service Codex, recently wrote on X: "I'm increasingly being asked during job interviews how much dedicated inference compute they will have to work with Codex." He added that usage per user is growing much faster than the overall growth of users, a sign that AI computing is becoming increasingly scarce and valuable.

This scarcity is redefining how engineers view their work and compensation. Greg Brockman made this clear: "The inference compute available to you will increasingly influence the overall productivity of software." In other words, access to AI could soon be as important as access to a high salary and stock rewards. As a developer in the AI era, if you do not have access to massive compute, you risk producing significantly less software than your peers, thereby threatening your career prospects.

Hakeem Shibly, a data specialist at Levels.fyi, recently spotted a compensation submission from a software engineer that included a "Copilot subscription" as part of the compensation and benefits, a small but symbolic step towards access to AI as a standard perk.

Towards AI Token Compensation

Some members of the AI community envision an even more explicit future. Peter Gostev, head of AI capabilities at Arena, a startup that measures model performance, stated: "OpenAI and Anthropic should create recruitment sites where their clients can post jobs, indicating the token budget for the position in addition to the salary range." These startups did not respond to requests for comment on Monday.

Investors are taking note. Tomasz Tunguz of Theory Ventures stated that companies are effectively adding AI inference as a fourth component of engineer compensation: salary, bonus, stock, and now tokens. Tokens are the economic language of generative AI. Models break down words and other inputs into digital tokens to facilitate processing and understanding. One token represents about ¾ of a word. They are also used to price the use of AI models, through an industry-standard cost per million tokens.

"Will you be paid in tokens? In 2026, you will probably start to be," said Tunguz.

Financial Implications for Companies

For CFOs, this potentially significant new expense must be monitored as closely as other workforce-related costs, Tunguz stated. "This is starting to happen," he said, as employee use of AI increasingly contributes to total cash consumption. "It's a consideration for the CFO's office."

With Levels.fyi estimating the salary of a software engineer at the 75th percentile to be $375,000, Tunguz estimates that adding $100,000 in annual inference costs brings the total cost to $475,000—meaning that over 20% of the compensation cost could come from AI usage in the future.

The key question for financial leaders: what is the return on this AI investment? If cloud infrastructure performance is judged by gross profit per hour of GPU usage, Tunguz suggests that the equivalent for employees is productive work per dollar of inference.

Tunguz has integrated AI tools and models into his daily workflow and automates 31 tasks per day at a cost of about $12,000 per year in inference. "The engineer who still spends $100,000? They must be 8 times more productive!" he wrote in a recent LinkedIn post.

If this trend continues, 2026 could mark the year when engineers negotiate not only their compensation in dollars and stock but also in tokens.