xAI cuts Grok Bot usage limits as Musk says costs outrun revenue

xAI has tightened the usage limits on Grok Bot, its always-on AI agent, without advance notice, and Elon Musk has confirmed the reason in a post on X: “The cost of serving Grok Bot was unsustainably high relative to revenue.” Subscribers found their weekly allowances exhausted far ahead of schedule over Halloween weekend, in some cases within hours, before any public announcement was made.

The episode is a clear, public example of a problem every builder of agent products faces: a flat monthly fee does not scale with the compute an autonomous agent consumes. This is the first Unicorn API article in the Grok Bot thread.

What happened to Grok Bot’s limits over Halloween weekend

According to TechCrunch, some Grok Bot users woke up to find their weekly usage limits unexpectedly maxed out, even if they had not been using the agent overnight. Users reported on X that usage jumped from as low as 50% to 100%. Gadget Review cites reports of usage going from 56% to 100% within a single hour, and Gizmodo quotes one user who wrote, “My usage went from 58% to 100% in less than an hour with nothing running. What is going on?” Another user told followers, “I was at 68% yesterday and woke up to 98%.”

Musk’s explanation came as a reply rather than an announcement. VARINDIA reports he was responding to a user posting as “DogecoinNorway,” who tagged him and asked whether Grok Bot limits could be reset because many people said their usage had run out very quickly during the week. Gizmodo dates the post to Sunday afternoon and describes it as a reply to a user who had hit the weekly limit earlier than usual and asked Musk personally for a reset. TechCrunch notes that Musk confirmed the change was not a glitch, and that he had not addressed the issue on his main feed while continuing to promote Grok Bot as a way to buy gifts for family.

Gadget Review reports that no advance public announcement was identified before the limits changed, and that no restoration timeline was offered. Citing TechSpot, it adds that when one user raised the issue directly, Musk reportedly asked whether upgrading to a higher-tier plan would be too expensive. Online frustration gathered around the phrase “indirect rug pull,” which TechSpot used to describe subscribers’ sense that a paid service had become less valuable without adequate notice.

How large the cut is remains unverified

The scale of the change rests on user reports. According to TechSpot, as cited by Gadget Review, some users claimed their effective limit had been cut roughly in half. Gadget Review stresses that these are user reports, not independently audited measurements, and that xAI has not publicly documented what the limits were before or after the change.

The outlets also frame the change slightly differently. TechCrunch calls it deliberate throttling. Gizmodo says Musk “appears to have imposed” a tighter-than-usual weekly token limit, and that it is unclear whether xAI will enforce a stricter limit permanently or whether this was a one-off emergency measure; xAI did not immediately respond to Gizmodo’s request for comment. Gadget Review says it remains unclear whether the reductions affected all three subscription tiers or only lower-cost plans.

Beyond Musk’s single sentence, the numbers that would settle the question are private. xAI has not disclosed Grok Bot’s operating costs, per-user cost, subscription revenue or profit margin.

How Grok Bot is sold and metered

Grok Bot is the always-on agent xAI launched in August. xAI promotes it as a kind of virtual teammate that handles complex tasks and checks back for human approval when needed. It runs on its own virtual computer rather than on the user’s device and can operate around the clock, even when the user is offline.

Access comes through X subscription plans, with higher tiers getting larger usage allowances. Subscribers receive a weekly “pool” of tokens (tokens are the units of text a model reads and writes, used here as the usage meter) that varies by plan, and they can buy “on-demand” credits that the agent uses once the weekly pool runs out.

Item	What the sources report
Premium plan	$30 per month
Super plan	$100 per month
Ultra plan	$200 per month
Metering	Weekly token pool by plan, plus optional on-demand credits
Where it runs	Its own virtual computer, always on
Launched	August
Paying users	About 800,000 weekly in early October

Gizmodo places Grok Bot in a field with OpenAI’s Dots and Meta’s Muse, which also run on their own virtual computers. Muse is free, which Gizmodo calls a big factor in its popularity, while Dots and Grok Bot are available only to paying customers.

Why agents strain flat-rate subscriptions

The cost pressure follows from how agents work. Unlike a standard chatbot reply, an agent may complete a task by making repeated model calls, browsing the web and taking computer actions in sequence; each extra step can add serving cost, while a fixed monthly fee does not scale with that usage. Gizmodo makes the same point: running continuously on a virtual computer makes these agents far more capable than limited chatbots, but they burn through tokens faster.

Growth made the problem more urgent. Citing TechSpot, Gadget Review reports that SpaceX President Gwynne Shotwell put Grok Bot at about 800,000 paying weekly users in early October, a figure that reportedly doubled the previous month’s count. VARINDIA also reports around 800,000 paid users and says the processing volume at that scale appears to have outstripped the subscription income meant to fund it. Musk had recently described Grok Bot as having “the fastest growth of a product that I’ve ever seen in my life.” As Gadget Review notes, rapid growth can worsen the problem if the cost of extra usage exceeds the extra revenue per user.

The timing has drawn comment. Gizmodo points out that, days before the cost post, Musk had written that “Saving money for the distant future is pointless with the advent of super intelligent abundance.” TechCrunch adds financial context: Grok Bot has been positioned as a success story for the newly public SpaceX, which said in a filing last month that AI makes up 52% of company income, and its renamed “Super Intelligence” division turned a profit of $431 million, which would have been higher without $1.2 billion in one-time costs tied to its acquisition of Cursor. The sources do not explain how that division-level profit squares with Grok Bot’s per-product economics.

An industry-wide pricing problem

The gap between flat-rate pricing and inference costs (the compute spent each time a model generates output) is not unique to xAI. Gadget Review cites a separate analysis, reported by TechSpot, estimating that fully using a $200 ChatGPT Pro subscription could correspond to about $14,000 in equivalent API charges; it stresses that this is a modelled comparison, not evidence of what OpenAI charges any subscriber. The same report puts OpenAI’s 2025 cost of revenue at roughly $7.5 billion, according to TechSpot. VARINDIA reports that rivals such as OpenAI are also said to be quietly restricting how much users can draw on their most advanced models to save money.

Analysis from Surf frames the competitive stakes: Musk’s statement suggests the current mix of pricing, usage and monetisation does not support the bot at scale, and AI competition is shifting toward cost per query, utilisation and revenue per user rather than model capability alone. Surf lists the likely responses as higher prices, tighter usage limits, more advertising or platform integration, and work on inference efficiency.

What it means for builders

For developers and founders shipping agent products, the episode carries several practical lessons drawn from what the sources describe:

• Price for the long tail of usage. Agents that loop through many model calls and actions break the cost assumptions of a fixed monthly fee. Usage-based components, like Grok Bot’s on-demand credits, are one way providers absorb heavy users.
• Background agents surprise users. Grok Bot users report allowances draining overnight or “with nothing running”. When an agent works while users are offline, a clear view of what consumed the budget matters.
• Communicate limit changes before they land. The backlash centred on the lack of notice and of published before-and-after limits, not only on the size of the cut. Gizmodo speculates that user complaints could carry legal risk, noting a class action that competitor Anthropic has faced.
• Do not assume stable consumer access. As Gadget Review puts it, a flat monthly price may not guarantee a stable level of access when the product performs expensive autonomous tasks. Teams relying on a consumer agent subscription for daily work may have to live within tighter caps until xAI settles its economics.

The sources discuss only the consumer Grok Bot subscription; none of them reports any change to xAI’s developer API. xAI’s separate API policy changes are covered in xAI’s Grok API will watermark text by default from November 7.

What to watch next

• Published limits. Whether xAI documents the new weekly token pools per tier, given that it has not stated the limits before or after the change.
• Permanent or temporary. Whether the stricter cap stays or was a one-off measure, which Gizmodo says is unclear, and whether Musk says how limits will work going forward.
• Pricing changes. VARINDIA reports it is unclear whether xAI will introduce new pricing, change limits further or streamline its backend to cut costs.
• Tier effects. Whether Premium, Super and Ultra subscribers were all affected, which remains unconfirmed.
• Rival moves. Whether OpenAI’s Dots or Meta’s free Muse face similar cost pressure as usage grows.