• Elon Musk confirmed SpaceX (SPCX) restricted Grok Bot usage over Halloween weekend, citing operating costs 'unsustainably high relative to revenue.'
  • The move follows rapid adoption: approximately 800,000 paid weekly users by early October, more than double the prior month.
  • SpaceX's AI segment generated $7.5–$7.6 billion in revenue last quarter, but the cost of serving the autonomous agent remains a pressure point.

Cost Pressures Force Limits

Elon Musk said on November 1 that serving the Grok Bot AI agent had become “unsustainably high relative to revenue,” confirming that SpaceX intentionally restricted usage over Halloween weekend. The disclosure came after subscribers complained that their weekly allowances were running out unusually quickly, with some reporting abrupt jumps in their usage meters — including a leap from 33% to 66% and accounts hitting their limits overnight.

The restrictions mark a sharp turn for a product that management had championed just weeks earlier. During October’s earnings discussion, SpaceX celebrated Grok Bot’s rapid growth, reporting approximately 800,000 paid weekly users in early October, more than double the preceding month. Grok Bot, an autonomous task-performing agent, is distinct from the ordinary Grok chatbot and is designed to execute complex, multi-step assignments.

Musk suggested that one dissatisfied user consider a higher-usage plan, but he disclosed neither Grok Bot’s standalone costs and revenue nor any definite pricing change or allowance reset. The specific mechanics of the throttling remain unclear, and it is not established whether every affected account was deliberately limited.

A Profitable Segment, A Costly Product

The restrictions have raised questions about the economics of AI agents, which consume far more computational resources than simple chatbot responses. Unlike a short query, an autonomous agent may repeatedly call models, browse websites and operate a computer before completing a single assignment. More delegation therefore means more inference — the computation used to run a model — and higher infrastructure expense, even when the subscriber’s monthly payment remains unchanged.

That tension is playing out against a backdrop of strong financial performance at the segment level. SpaceX’s AI operations, presented under the “Super Intelligence” banner, generated approximately $7.5–$7.6 billion in revenue last quarter, or about 52% of group revenue, with operating profit of $431 million. Those results included $1.2 billion in acquisition-related stock compensation expense. Group revenue for the third quarter reached $14.6 billion, with net income of $4.4 billion.

The important distinction is that the 52% figure refers to revenue, not profit. Likewise, the $431 million operating profit belongs to the broader AI segment — which also includes the Cursor coding assistant and data-center capacity rented to customers — not Grok Bot alone. That segment’s profitability does not contradict Musk’s statement that one particular product costs too much to serve.

“What we are focused on is making these agents economical to operate at scale,” Musk said during the earnings call, according to people familiar with the matter. He also told investors that power, cooling and infrastructure, rather than chips alone, were becoming the principal deployment constraints, describing turbine-component supply delays of 12–18 months.

Competitive and Regulatory Crosscurrents

Usage rationing is not unique to SpaceX. Anthropic’s current pricing explicitly combines weekly allowances with plan-dependent session limits, and consumption depends on factors such as model choice, conversation length and file size. On October 30, Anthropic removed the five-hour session cap for verified users of its $200 Ultra plan while retaining the weekly allowance — illustrating the same industry trade-off of allowing longer uninterrupted work without offering unlimited total computation.

Regulatory issues are also in play. In January, reports emerged that an EPA rule update clarified that qualifying data-center gas turbines require Clean Air Act permits and cannot simply be treated as temporary “non-road engines,” potentially affecting the speed and cost of AI infrastructure expansion. On October 31, a federal judge allowed an NAACP lawsuit concerning allegedly unpermitted turbines at an xAI facility in the Memphis area to proceed — a procedural decision, not a final determination of liability.

SpaceX did not respond to a request for comment on the weekend restrictions or on the status of affected users’ allowances.

What Comes Next

Tighter quotas, more differentiated subscription tiers or additional paid usage are plausible responses, but they remain scenarios rather than announced Grok Bot policy. Musk’s comments establish the cost problem; they do not establish the next pricing decision, compensation for affected users or a timetable for restoring access.

The key unanswered questions are the actual allowance change, the cost per completed task, and whether lighter and heavy users can be served profitably under the same subscription structure. Current reporting notes that the necessary product-level financial figures remain undisclosed.

Longer term, the company faces two different tests: making agents economical to operate and financing the infrastructure needed to scale them. Management remains highly optimistic: CFO Bret Johnsen forecast $33–$37 billion in group revenue for the first quarter of 2027, while Musk projected annual recurring revenue of $500 billion to $1 trillion by the end of 2027. Those are management forecasts, not independently established expert consensus, and should be weighed against the group’s year-to-date loss of $413 million over the first nine months and enormous planned capital spending — $19.6 billion in the third quarter alone, with a forecast of $35–$40 billion for the fourth quarter.

The most defensible interpretation is narrower than “AI is unprofitable”: SpaceX reports a profitable AI segment, while its fast-growing personal agent is exposing the difficulty of matching fixed subscription revenue to variable computing costs.