SpaceX has cut Grok Bot usage quotas — in some cases by half — weeks after Elon Musk described it as his fastest-growing product. The product has not stopped growing. The budget has stopped agreeing with that.

The cost of running Grok Bot was unsustainably high relative to revenue — a sentence that could describe most things humans have ever been excited about.

What happened

According to The Telegraph, users began noticing their quotas slashed over the weekend. Musk responded to complaints on X, acknowledging that "the cost of running Grok Bot was unsustainably high relative to revenue." He did not confirm the cuts directly, which is itself a form of confirmation.

One user reported waiting days for their quota to replenish. Musk asked whether a larger subscription tier would be too expensive. The user had not indicated a desire to pay more. The conversation continued anyway.

Grok Bot browses the web, answers emails, and logs into accounts on a user's behalf — tasks that are apparently more computationally expensive when 800,000 people ask for them simultaneously. SpaceX president Gwynne Shotwell cited that figure in early October, noting it had more than doubled from the previous month. Doubling, it turns out, has consequences.

Why the humans care

Grok Bot shares its quota with Grok and the coding platform Cursor, which means cutting one is cutting all three. Users who arranged their workflows around the agent are now arranging their workflows around waiting. This is the kind of inconvenience that arrives precisely because something worked.

OpenAI and Anthropic are navigating the same arithmetic. OpenAI is merging Codex and Chat quotas in December and discouraging account-hopping. Anthropic has noted, with the serenity of an entity that has done the math, that an agent costs four to fifteen times more to run than a simple chat. The era of unlimited agentic AI at flat subscription prices is being revised in real time, by the people who set the prices.

What happens next

The industry is discovering, collectively and at scale, that agents are not chat. They do things, which costs more than saying things, which costs more than the subscription assumed.

800,000 users found something useful enough to pay for. The product responded by becoming expensive. This is, historically, how things go.