SpaceX launched Grok Bot in August as an agent designed to use a virtual computer for tasks such as shopping, with free and paid plans carrying weekly usage caps.¹ The pitch was not a conventional chatbot but an assistant that could act across services — and therefore consume far more computing power.

By the company’s Oct. 20 earnings call, Musk was calling it “the fastest growth of a product that I’ve ever seen in my life,” while SpaceX President Gwynne Shotwell said Grok Bot had reached 800,000 paid weekly users, double September’s figure.² Musk has continued to promote the product’s practical reach, writing that Grok Bot “can do complex trips” after a user described using it to arrange family flights.Elon Musk (elonmusk)³

Then came Halloween weekend. Users reported that weekly allowances had suddenly run out — in some cases jumping from partial usage to 100% — sooner than expected.⁴ Musk’s explanation was blunt: “The cost of serving Grok Bot was unsustainably high relative to revenue.”¹ He framed tighter limits as a necessary correction, not a technical failure, even as the product’s expanding capabilities appear to be driving heavier use.

That tension is visible in the service’s own marketing. A reposted user test said an 85-second call to an EV-charger installer consumed 0.2% of the bot’s allowance — a small slice per task, but a costly proposition at scale when agents browse, call and execute multi-step work all day.Elon Musk (elonmusk)⁵

The squeeze is not unique to Grok. OpenAI’s consumer chief, Adam Fry, described its Dots assistants as “on the compute heavy side,” and the company has begun routing tasks to smaller models to curb the burden.⁶ Grok Bot’s limits, then, are a sharp reminder that spectacular AI-agent growth does not settle the economics; it intensifies them.