0xMarioNawfal

SOS 🆘 Grok Bot users had a rough week 💀 Many of them burned through 100% of their weekly usage days before the reset, including people who say they never hit the limit. One user watched his meter leap from 44% to 87% in just 24 minutes. When one of them asked Elon for a reset, he didn’t hand one out. Instead, he got refreshingly honest and said the cost of serving Grok Bot was unsustainably high relative to revenue 😱 That one line tells you a lot. Notice what he didn’t say. He didn’t blame a shortage of chips. He said the bot costs more to run than it earns. That turns this into a question about money, not hardware: what else could those chips be earning right now? Grok Bot is a hungry product. It isn’t a chatbot sitting around waiting for your next question. It’s an agent with its own cloud computer, and it keeps grinding through tasks long after you close the app. SpaceX says it hit 800,000 paid weekly users in October, more than doubling in a single month. Even more striking, people use it more every week, not less. Yet it’s sold on flat monthly plans, with the main X Subscription tiers at $30, $100 and $200. That’s an all-you-can-eat buffet price for customers whose appetites keep growing. Here’s where it gets interesting. On that same earnings call, SpaceX revealed that new compute pays for itself in under a year at today’s rental prices. Elon said the goal is to split compute roughly 50-50 between SpaceX’s own products and outside customers. SpaceX has even pushed some of its inference work out to other cloud providers so it can keep more of its own chips for training. Think about what that means. Every GPU hour Grok Bot burns is an hour SpaceX could have rented out at market rates, and those rates keep climbing. According to SemiAnalysis, a one-year H100 rental contract jumped almost 40% between October 2025 and March 2026. That chip came out in 2022. Hardware is supposed to get cheaper as it ages, yet this one got more expensive. Now run the numbers on a newer Blackwell B200, which rents for around $6 an hour. At that rate, a $30 subscription buys roughly five hours of one B200 per month. To be fair, a single bot doesn’t tie up a whole GPU, since servers share each chip across many users at once. But an agent working all week chews through those five hours far faster than any chat window ever could. To me, this is a pricing problem far more than a demand problem. Only 4.5% of US consumers paid for a personal ChatGPT, Gemini or Claude subscription as of August, but that share doubled in just one year. The appetite is clearly there. In Elon’s replies, some users are asking for $500 and $1,000 tiers, while others want a cheaper, lighter model for basic tasks. Both camps want the same thing: pricing tied to how much work the bot actually does. My bet is that AI agent pricing will start looking a lot less like Netflix and a lot more like your electric bill. And for SpaceX, that’s a fantastic problem to have. When the fastest-growing product Elon has ever seen can’t cover the cost of running on SpaceX’s own chips, it tells you just how valuable those chips have become.

Anthropic ($ANT) Pre-IPO Perp is now live on Polymarket Perps Up to 10x leverage. No market hours. No expiry.