Grok Bot ran out for a wave of paying users this week, and Musk didn’t blame demand. He blamed the bill.
A user on X asked Elon Musk on November 1 whether xAI could reset Grok Bot’s weekly usage pool, saying they and plenty of others had burned through it far faster than usual. Musk’s answer skipped the usual we’re-scaling-to-meet-demand line. “The cost of serving Grok Bot was unsustainably high relative to revenue,” he wrote. No dollar figure, no timeline for a fix, just the admission that the product was losing money on every heavy session.
That’s a different kind of confession than the AI industry usually makes in public. Labs love to talk about demand outstripping supply. They rarely say, in plain words, that the thing people are using too much of is the thing that’s bleeding them.
Grok Bot isn’t a chat window. xAI launched it on August 11, 2026, as an agent that runs tasks on its own virtual computer in the background: research, writing, browsing the web, multi-step jobs that keep going after you close the tab. That’s expensive by design. A single agent run can chew through far more GPU time than a one-off chat reply, because the model is reasoning, searching, and executing in loops rather than returning one answer and stopping.
xAI never gave Grok Bot a standalone price. It rides along with paid subscriptions, SuperGrok at $30 a month and up, or through Cursor plans, each carrying its own weekly allowance of agent compute that xAI has never published as a hard number. There’s no free tier for the agent itself. According to reporting from Teslanorth, the product grew to roughly 800,000 paid weekly users within about two months of launch, a fast ramp by any measure. The problem Musk just admitted is that fast growth didn’t translate into money that covered the GPU bill.
This isn’t the first time Musk has said as much, just more bluntly this time. Earlier this year he floated adding ads inside Grok’s answers specifically, in his own words, to help pay for “those expensive GPUs.” That was a plan for the future. This week’s comment was an admission about the present: the serving cost on Grok Bot right now doesn’t clear what xAI is pulling in from it.
A tell the rest of the industry has been dancing around
OpenAI has already started metering token usage more tightly across its paid tiers this year, a quieter version of the same math problem. xAI’s version is louder because Musk said it out loud, on the record, with no hedge. Frankly, that’s rare. Most AI companies would rather talk about scaling infrastructure than admit a flagship feature is underwater at current prices.
It also complicates the pitch xAI has been making to investors and to the Pentagon, which signed a $200 million deal with xAI for Grok access. An agent product that can’t cover its own compute costs at 800,000 weekly users is a product that needs either a price increase, a much cheaper inference stack, or a subsidy from elsewhere in the business. Musk didn’t say which. He didn’t announce a shutdown, a price hike, or a fix. He just confirmed the number doesn’t work yet.
Inference costs, not interest, are what capped Grok Bot’s usage this week.
Also read: Andrej Karpathy says 6,454 Hacker News picks beat YouTube’s algorithm for him • AI Data Centers Are Draining the RAM Supply and Your Next Laptop Pays For It • How to Negotiate an AI Agent Training Data Indemnity Clause
This article is posted in AI News, check it out for more related stories.
Almost there. Sign in and your reply posts straight away.
