Elon Musk is pushing Grok Bot users toward a feature that, until this week, few of them could actually use. “Try asking Grok @Bot to make calls!” Musk wrote on X Tuesday afternoon, quoting a demo from Sawyer Merritt that included audio from a live call.

Musk first announced calling on October 25, writing that the agent could “call someone or a business for you, make a request on your behalf, take notes on the response and send you a note summarizing the conversation.” There were no release notes at the time, and the feature only appeared in preview builds. That changed on Tuesday, when creator Peter Yang said all Grok Bot users should now be able to call US businesses. Musk reposted his demo.

Try asking Grok @Bot to make calls! — Elon Musk (@elonmusk) November 3, 2026

Yang’s call shows how far the feature goes. In his video, his bot dialed Comcast, worked through the phone menu, deflected small talk about the weather by saying it does not “have eyes on the sky,” turned down three upsells and used a cheaper fiber offer from competitor Sonic as leverage. His bill dropped from $84 to $45 per month on a faster plan, a savings of nearly $500 a year. The call ran about 25 minutes, and the bot paused to get Yang’s approval before accepting the final terms.

Merritt described what the user sees: a live transcript runs during the call, and when the business asks for something the bot does not know, an answer box pops up in the chat. The user types a reply and Grok relays it to the person on the line. Other users posted results Tuesday, including a $100 gift card recovered from Verizon Fios after a 51 minute call and a same day pediatric appointment, which Musk also shared. Not every call worked. One user said a rep refused to cancel a subscription once it was clear an AI was on the phone.

Grok @Bot can now call someone or a business for you, make a request on your behalf, take notes on the response and send you a note summarizing the conversation! — Elon Musk (@elonmusk) October 25, 2026

The Grok Bot terms require every call to identify itself as a virtual assistant calling on the user’s behalf and warn that calls may be recorded. Marketing, mass calling and impersonation are prohibited, and calling currently works only with US numbers.

The timing stands out. Days ago, users began hitting their weekly limits early, and Musk said the cost of serving Grok Bot was “unsustainably high relative to revenue,” as Teslarati reported. A call that spends most of its 25 or 50 minutes on hold is the kind of long running task that eats into usage, and SpaceXSI has not said how calls count against the weekly pool. Grok Bot is included from the $30 Premium tier of The X Subscription upward.

For Tesla owners, the agent has been in their cars since September 22 and is enabled across every Cybercab and Model Y Robotaxi. Musk has already pitched it for phone chores like scrubbing your number from spam call lists. Neither Tesla nor SpaceXSI has confirmed whether a bot can place one of these calls on request from inside the vehicle.

Elon Musk said weeks ago during SpaceX’s third-quarter earnings call that the company’s Grok Bot, its autonomous AI Agent platform, was showing “the fastest growth of a product” that he’d ever seen in his life.

Now, SpaceXSI is scaling back Grok Bot users’ weekly limits, much to the surprise of those who have relied on it to take away monotonous tasks in their lives.

Elon Musk’s terse November 1 reply on X captured a sharp tension in the early days of agentic AI. Responding to a user who asked for a Grok Bot usage reset after hitting 100 percent with four days still left on the weekly cycle, Musk wrote: “The cost of serving Grok Bot was unsustainably high relative to revenue.”

@DogecoinNorway The cost of serving Grok Bot was unsustainably high relative to revenue

— Elon Musk (@elonmusk) November 1, 2026

The post arrived amid a wave of similar complaints from paying subscribers whose meters had jumped unexpectedly toward or to the limit, sometimes overnight and with little corresponding activity. It also stood in clear contrast to statements Musk had made only days and weeks earlier.

On the SpaceX third-quarter earnings call in mid-to-late October, Musk had described Grok Bot in almost the opposite terms. He called it “the fastest growth of a product that I’ve ever seen in my life” and said he had “never seen anything grow this fast.”

Company figures put paid weekly users at roughly 800,000 earlier in the month, more than doubling over the prior period. Daily active users had climbed from under 100,000 in early September to over 600,000 by late October.

Musk emphasized that the product has no free tier and that users appear to be getting real value beyond what they pay. In other remarks, he noted that usage per user was rising rather than declining, the opposite of the usual decay pattern for new software.

Grok Bot, launched around August, functions as an always-available agent with its own persistent cloud computer; it can schedule, research, shop, browse, and execute multi-step workflows even after the user steps away. Access is bundled into X subscription tiers (roughly $30, $100, and $200 per month) and certain Cursor plans, drawing from shared weekly usage pools whose exact sizes have not been publicly detailed.

The economics of that model are what Musk’s November 1 post put in the spotlight. Agentic systems burn far more compute than ordinary chatbot exchanges. Each task can trigger repeated model calls, tool use, web actions, and sustained background work.

When growth is as rapid as Musk celebrated, the absolute compute bill scales quickly even if average revenue per user stays fixed under flat monthly subscriptions. Industry observers noted the opportunity cost as well: the same GPUs can be rented to other AI labs at higher rates.

A simple back-of-the-envelope comparison circulating after the post illustrated the gap: if every reported daily user paid the top tier, annualized subscription revenue would still be modest relative to the scale of compute infrastructure SpaceX is building and leasing to companies like Anthropic.

Some users also reported possible dashboard anomalies around the same period, though Musk’s framing treated the underlying cost-revenue imbalance as the core issue rather than a pure metering glitch.

The juxtaposition is the core of this exchange. Within roughly two weeks of publicly framing Grok Bot as an unprecedented success with strong unit economics and no free tier, Musk acknowledged that serving the same product had become unsustainable relative to the revenue it generates.

For the broader industry, the episode underscores a structural challenge of the agent era: demand is real and growing faster than many expected, yet the cost of fulfilling that demand through always-on, multi-step agents can outrun subscription pricing before companies can adjust limits, introduce metered billing, or raise prices.

Musk’s reply did not promise an immediate reset or higher allowances; it simply stated the constraint. In doing so, it supplied a rare, explicit public data point on the unit economics of a high-profile AI agent just as adoption numbers were still being celebrated as the fastest in Musk’s experience.

The tension between rapid, paid growth and the compute required to deliver it is now on the record.

Elon Musk is pushing back on one of the biggest worries hanging over SpaceX stock. In a reply on X on Sunday, Musk said the company’s capital expenditure efficiency is severely underestimated. He was responding to a Milk Road Stocks post citing analyst Kyle Reidhead, who estimates SpaceX will need to spend roughly $470 billion over the next five quarters.

Reidhead built that figure around the target SpaceX set on its third quarter earnings call: 13 gigawatts or more of compute by the end of 2027. In his breakdown, Nvidia captures 83 to 89 percent of the spending. SpaceX ended the quarter with 1.7 GW of nameplate compute and guided to $35 billion to $40 billion in fourth quarter capex, as Teslarati reported.

Musk did not attach a number to his claim, but the earnings call shows the argument. CFO Bret Johnsen said new compute currently pays for itself in under a year at today’s pricing, and Musk said that window could shrink to six months next year. Using that pace, Reidhead argues about $230 billion of the $470 billion could come from operating cash flow and about $60 billion from SpaceX’s roughly $90 billion cash pile, leaving around $180 billion to borrow. Musk confirmed last week that SpaceX had already closed a new $40 billion debt raise to buy GPUs.

The ground buildout is not the end goal. It is what is meant to pay for the move to orbit. Replying on Friday to charts built from Sequoia partner Shaun Maguire’s SpaceX model, Musk wrote that “Terrestrial compute will be peanuts compared to orbital compute,” adding that orbital compute deployed per month “probably exceeds terrestrial around the end of 2028 and then far exceeds it in 2029.”

Maguire’s model shows how steep that path is. It has capex climbing from $88 billion this year to $290 billion in 2027 and $369 billion in 2028, with SpaceX burning $77.8 billion in cash this year and $122.6 billion next year before turning $114.9 billion cash positive in 2028. Musk’s response to the model was, “I think we can do better.”

The orbital half runs through Starmind. SpaceX plans to operate up to 1 million AI satellites flying in clusters of about 10, linked by roughly 10 terabits of bandwidth, with deployment of thousands of units starting as soon as late 2027 out of its Gigasat factory in Bastrop, Texas, according to SpaceX’s space safety page. Musk has said each Starmind spacecraft will carry more than 250 kW of solar power.

Getting there depends on launch volume. Maguire’s model assumes 250 Starship flights in 2028, up from two this year and 38 next year. The nearer test is financial: SpaceX guided to $33 billion to $37 billion in first quarter 2027 revenue, more than double the $14.58 billion it booked in Q3, and Musk reposted that guidance again on Sunday morning.