Model

xAI ships AI teammates at $30-$200 a month, ten days after Musk said serving them loses money

The published card sets a chatbot price against an always-on machine the CEO says loses money on every subscriber; SpaceX shareholders fund both sides of that trade in one stock.

Richard TangRichard Tangfour-minute read
Share
Elon Musk walking through the venue of the UK government's AI Safety Summit at Bletchley Park in November 2023, in conversation with a delegate
1 / 6Slide 1 of 6
Elon Musk at the AI Safety Summit at Bletchley Park in November 2023. Ten days before Grok Bot 1.0 shipped, his post on X conceded that serving Grok Bot cost more than its subscriptions brought in.

xAI$200B — xAI, private, latest valuation $200B shipped Grok Bot 1.0 on November 11: AI teammates that sit inside Slack, Microsoft−0.28% — Microsoft, down 0.28 percent today Teams, Zoom+1.64% — Zoom, up 1.64 percent today and Google Meet, each running on its own cloud computer around the clock, sold inside subscriptions from $30 to $200 a month 12. The design carries Cursor's playbook, the collaboration product xAI paid about $60 billion for, from code into office work: a team trains one shared bot, and the knowledge stays when people change roles 1.

Ten days before launch, Musk said the price did not cover the cost

Ten days earlier, Elon Musk conceded the assumption the price card rests on does not hold. "The cost of serving Grok Bot was unsustainably high relative to revenue," he wrote on X on November 1, after subscribers on the $100 and $200 tiers drained their weekly allowances over the Halloween weekend and were locked out until reset 34. Our November 10 piece on Grok 4.8's missing price card said the card was the next receipt; this is what it showed (/article/xai-s-grok-4-8-entered-training-57-days-late-with-no-price-card-anthropic-pays-spacex-1-25-billion-a-month) 34.

The stickiest features are the costliest

The overspend is architectural, not wasteful. A plain chatbot routes a question to inference and back; an agent that works all day on a dedicated machine runs at roughly one CPU per GPU, against one per four to eight for chat, by TD Cowen's comparison of the two architectures 1. Grok Bot's retention features, the shared bots and the meeting attendance, are its heaviest consumers of compute 12.

The same bill lands at Meta

They are working: Gwynne Shotwell, chief operating officer of xAI owner SpaceX−2.57% — SpaceX, down 2.57 percent today, told investors on October 20 that Grok Bot had 800,000 paid weekly users, more than doubling in a month, with companies using it as an always-on teammate across sales, marketing, IT and engineering 5. Musk added that it may end up consuming most of SpaceX's Super Intelligence compute, the company's pooled capacity for training and running its AI models 5. Nor is this an xAI execution slip. Mark Zuckerberg told investors Meta's+2.23% — Meta, up 2.23 percent today rival Muse agent is "very compute-intensive" and not yet profitable 3. Sell a 24/7 teammate at a chatbot price and you are selling compute below cost 3.

One balance sheet funds both sides

SpaceX shareholders own that trade. Third-quarter revenue reached $14.6 billion, and the AI business passed half of it 57. Capital spending nearly tripled year on year, and the chief financial officer guided the December quarter higher still; on November 12 SpaceX added $74 billion of new campuses in Texas and Arizona 57.

The tenants with signed sums are Anthropic$350B — Anthropic, private, latest valuation $350B, up to $120.5 billion through 2029, and Google−1.54% — Google, down 1.54 percent today at $920 million a month 7. Musk's own stated plan is to rent about half the compute and keep the rest for internal use 5. On that split, the rented half arrives as fixed contracted revenue while the half handed to Grok Bot exceeds its subscription revenue, on Musk's own words.

SpaceX capex now runs $5 billion a quarter ahead of revenue

  • Revenue
  • Capital expenditure
  • Estimate
$0B$10B$20B$30B$40BQ1 '25Q2 '25Q3 '25Q1 '26Q2 '26Q3 '26Q4 '26 guided$35–40B$19.59B
Data
RevenueCapital expenditure
Q1 '25$4.07B$4.14B
Q2 '25$4.07B$2.83B
Q3 '25$5.58B$6.88B
Q1 '26$4.69B$10.11B
Q2 '26$7.81B$18.37B
Q3 '26$14.58B$19.59B
Q4 '26 guided (estimate)—$35B ($35–40B)
SpaceX revenue and capital expenditure by quarter, $ billions. The Q4 2026 bar is the low end of the company's guided $35 billion to $40 billion capex range; Q4 2025 is omitted because those figures were not reported. Sources: SpaceX quarterly figures from its SEC filings; Q3 2026 earnings call.6,5

The meter is the broken link

Team memory is a durable product bet: shared context is what made Cursor$30B — Cursor, private, latest valuation $30B a business while chat assistants stayed a commodity, and the team that routes work through a shared agent does not go back to doing it by hand. The party that benefits first is whoever owns the memory; today Grok Bot's shared knowledge belongs to the employee who trained it, not the company that pays 1.

An enterprise will not route real workflows into an asset it does not hold. The broken link is the meter, and two fixes decide whether the bet pays: a price that covers what a 24/7 dedicated machine consumes, and team knowledge owned by the company rather than the person 1.

Three receipts will settle the bet

The next receipts are any separate line for agent revenue in SpaceX's fourth-quarter filing, the tenant commitments behind the $74 billion, and Anthropic's IPO pricing on November 18 78. The product bet is sound; until the meter matches the machine, the compute bill is being paid by the data center buildout rather than the subscribers.

Get Model by email, free

The AI labs, their models, and the business behind them. In our daily email, every weekday.

Loading SPCX

Reader comments

0 comments

    Model

    Google ships Gemini 4 Hydrogen at $1.50 a million tokens and gives it away in Search

    Gemini 4 Hydrogen scores 51 on the Artificial Analysis index against 50 for the flagship Argon, at $1.50 per million input tokens with no long-context surcharge, and it is free in Search. That strips both halves of a model lab's premium case the same week Anthropic prices an IPO on one.

    Model

    Scale AI just published the benchmark that prices its own Qatar contracts

    Scale AI's new Singapore legal benchmark measures exactly the gap that its Qatar contracts, signed three days earlier, pay Scale to close. Oracle's sovereign-AI story and private legal-AI valuations now have a number to argue with.

    Model

    Suno bought the top AI music spot with licenses; ACE Music v1 just took it, 1,160 to 1,144

    Suno bought the top of the AI music chart with licensed Warner, BMG and Believe catalogs; six weeks later a challenger holds it, and that challenger's model family is going free. The pricing test lands on Spotify's planned $4 to $5 remix tier and on what the majors can charge for AI training data.

    Model

    Midjourney says V9 could double its cost per image

    The self-funded lab is asking subscribers to accept twice the per-image cost for quality, the same week MiniMax open-sourced a 4K model whose license skips the US and the EU. The answer will price every credit-metered image AI, Adobe's included.

    Model

    Four video apps switched on ByteDance's 4K model and none named a price

    ByteDance's flagship video model reached its first consumer shelves on November 11, and the same day an API reseller cut its price below the model owner's own list. Who keeps the margin on a second of AI video is the open question, and the publicly traded read is Adobe.

    Model

    Google prices Gemini Decisions at 10 cents a million tokens, with no output charge

    Bounded agent work, the routing and grading calls every AI system makes thousands of times a day, is being priced toward zero by the two biggest distribution owners in AI, days before Anthropic prices an IPO on premium tokens.

    Subscribe to TheInference for free