Grok Bot is too expensive to serve: Musk admits costs exceed revenue

By Nate (Nate's Substack) · November 4, 2026.

The email starts from a statement attributed to Elon Musk: the cost of serving an AI agent, Grok Bot, is too high for what it brings in. According to the text, a Grok Bot customer had used up their quota four days before the reset and asked for it to be restored early. Musk's explanation, quoted verbatim in the email, was: "The cost of serving Grok Bot was unsustainably high relative to revenue."

The author sets that claim against a figure from xAI itself. In a "procurement write-up" published in September, xAI explained that Grok Bot found 43 paid licenses for a software product that had seen no activity in 90 days. The review and the license downgrades that followed saved $14,220.

Nate's central thesis is that both things can be true at once: an agent can create substantial value for its user and, at the same time, cost the provider too much to serve at the current price. According to the author, the cost of doing the work, the price charged for doing it and the value of having it done are three different figures, and we will have to learn to reason about all three much better.

The underlying question he raises is what happens when an individual can afford to direct work that used to require assembling a team. Nate sees this as one of the biggest changes agents bring to everyday working life. More and more people will decide for themselves what work to fund, how much effort it deserves and whether the result justified the expense.

The author describes two possible scenarios. On one hand, people can explore more possibilities, build things that would otherwise stay in a notebook, or ask for help with a problem before it becomes urgent enough to justify other people's time. On the other, they can spend a lot of money generating work nobody needs and then spend their evenings reviewing it.

The author notes that in the video accompanying the piece he promised to help calculate what one's own time and agents' work are worth. He says the article will do so and will include an interview readers can give their AI to analyze their own situation. The listed content is as follows: - Reopening work people had stopped considering, meaning ideas shelved because the work they required made them impractical, and seeing how an agent changes the cost of exploring them. - Setting a working range for the value of one's own hour, valuing time by its next-best use and testing everyday decisions against a range such as $20 to $50. - Running the numbers on an inherited weekly report, showing how the setup pays for itself in the second week and why a $20 option can beat a $5 one once attention is counted. - Budgeting an untested idea by choosing the maximum money and human attention one is willing to spend to find out whether people would use it. - Having the AI interview the reader, one question at a time, ending with a decision note and a briefing for the agent that will do the work.

The email cuts off there. The rest of the post, the Nate's Library MCP and the Slack community are for paid subscribers, so the full calculations and conclusions are not available.