Two weeks ago, Elon Musk was celebrating Grok Bot as the fastest-growing product he'd ever seen. On November 1, he acknowledged a problem: "The cost of serving Grok Bot was unsustainably high relative to revenue." The admission came after paying subscribers complained that their weekly usage allowances were suddenly exhausted, sometimes days earlier than expected. And this wasn't a product struggling to attract customers.
According to SpaceX's October earnings call:
- Grok Bot had approximately 800,000 paid weekly users.
- That figure had more than doubled since September.
- Customers were using the service more frequently as its capabilities improved.
That's the paradox: The more successful the product becomes, the worse its economics may get. Traditional software has enormous upfront development costs but relatively low costs for each additional user.
AI agents are different. Every time an agent browses websites, writes code, processes documents, or completes a multistep task, it consumes additional computing resources. And unlike a conventional chatbot, an autonomous agent might keep working for hours. A $30 or $100 monthly subscription looks attractive until a power user consumes more computing resources than the subscription generates in revenue.
To be clear, Musk hasn't disclosed Grok Bot's actual costs, revenue, or profit margins. SpaceX's broader AI division reported an operating profit last quarter, so this isn't proof the entire business is losing money. But it raises an uncomfortable question about the AI investment boom.
Companies are spending billions building infrastructure to support increasingly capable agents. What happens if customers love those agents but aren't willing to pay what they actually cost to operate?
Maybe efficiency improvements solve the problem. Maybe subscriptions get substantially more expensive. Or maybe AI companies start charging for actual work performed instead of offering generous flat-rate plans. The demand appears real. The sustainable pricing model is still being figured out.
Would you pay $200 a month for an AI agent that genuinely saves you 20 hours of work, or will AI companies have to keep subsidizing usage to maintain their growth?
https://x.com/elonmusk/status/2116992258520064397
https://fortune.com/2026/11/03/grok-bot-is-growing-fast-but-elon-musk-says-its-costs-are-unsustainably-high-compared-to-revenue/
https://twiiit.com/elonmusk/status/2116992258520064397
The tightened limits haven't had that big of an impact in my experience. The biggest grievance is weekly limits instead of monthly which aren't great for bursty development.
Tightening it cuts the tail of users off that probably generate a bunch of waste tokens.
Bot is about flows and customization, so there's a significant lock-in effect to it. Practically giving it away at first is how they quickly got scale, now they can trim. If they lost 10% of the 800k but saved 40% on capacity then that'a huge change to the economics.
SpaceX is unique too in that it doesn't have to worry about stranded compute, it's already printing big off google and Anthropic with any resources it doesnt use, so Bot has to be (or should ideally be) more profitable per watt than leasing out.
What they did that really sucked for users is the X unification account, I made that mistake. Doing so rugpulled external model credits, which were most of the value of the "Pro" and higher plans.
It sucks as a user, but good as a shareholder. I downgraded my plan, so less top-line revenue for them, but now none of my spend with them is going to external providers that was probably creating negative margin.