# The user's story — verbatim

1. GPUs are capacity - if they're idle price plummets - like an airline seat or deepwater drilling rigs. 
2. The current demand outlook looks soft, despite INCREDIBLE growth, greath results from Meta and Tesla are seen as *bearish* near term until *real, economic* use cases come online (e.g. drones, humanoid robots). 
3. Anthropic's leaked S1 has great growth that is slowing. Openrouter OVERALL token usage is slowing. OpenAI's revenue was $50B vs. $65B predicted. None of this is BAD but below expectations. 
4. All of the compute is *prepaid* and *contracted* and *delayed* in many critical areas. But this means you need to fund the payments with customer revenue. This seems unlikely: tokens (companies will wait 6 months for cheaper tokens, the biggest spenders are optimizing toward lower end models)  and new use cases aren't coming fast enough. 
5. Current usage is driven by "Free" agent use (dots, duet, grokbot, google search). If models were so constrained that they didn't have enough compute, they'd be selling these tokens not giving them away proactively. 
6. The real use case seems to be a 'rush to research' in biotech and other areas, but that is basically *brains without legs*; even if you achieve RSI, your implementation timeline would be long and revenue wouldn't be unlimited.
7. This is SUPER bullish long term and the companies are making the right bets, but they're trapped into overspending and perhaps insolvency. It would take a miracle that could replace workers MAJORLY in the next 6-9 months at large scale that would save the economics of the business, and it's not clear that will be manageable socially.
8. But near term, you can imagine a crash in token prices and many players weakening and stock prices plummeting as a result.
9. There simply are NOT ENOUGH use cases. The single one that has massive economic value (programming) seems to be a cost *driver* not a cost saver. Quality isn't there to ship 'straight to production' so it's just creating larger surface areas. Core products like Codex still ship with awful bugs (taking up disk space, e.g.) and Sam Altman, the quintessential user said in a YouTube video he only saves 1 hour per day with an ULTRAFAST DOT in a PURELY AI LEGIBLE COMPANY. This is bad. Legal, all these other companies will catch up and they're trying with RL environments and more compute, but it's not obviously they'll get there in time.
10. Most important it's not showing up in the main customer's P&L - to grow the economy by 10-20% you either need to produce the trillions in revenue or provide it to the people. This according to our MASSIVE orangebook analysis isn't happening.