AI 投入量波动

Rohan Paul (@rohanpaul_ai) 转发了 Aakash Gupta (@aakashgupta) 的帖子： Somewhere out there is a single user burning $35,000 a month in AI tokens, and that one number explains why enterprises are suddenly rationing models. Run it backward. At premium pricing of roughly $15-30 per million tokens, $35K a month is over a billion tokens from one seat. No human types that. That number only shows up when someone points agents at a problem and lets them loop. Usage stopped tracking how fast people work and started tracking how fast machines do. That detail breaks every pricing model sitting underneath it. Seat-based and flat-rate plans assumed human-paced consumption. Teams blowing past quotas by 200% means real usage landed at 3x what anyone budgeted, because the budget was set for a world where a person sat between the keyboard and the API. So enterprises did the rational thing and metered it. The moment you meter tokens, you start asking which ones actually need a frontier model. For most queries the answer is none of them. Summarize this, classify that, draft the email, rewrite the paragraph. A Qwen or DeepSeek model running at pennies on the dollar clears those at the same quality. That is what model routing really is. Hard reasoning, long context, and real code stay on the premium tier. Everything else drops to whatever runs cheapest and competently, increasingly open weights on rented GPUs. The volume lived at the bottom of that stack. The margin lived at the top. The two just got split apart. The labs keep the queries hard enough to justify premium tokens, and that band is real and defensible. It is also far narrower than every AI call an enterprise makes. Aggregate enterprise AI spend is still climbing. Whether that spend reaches the labs or clears on a cheap open model is the entire fight, and it's the part the 60% headline buries. > **引用原帖 Rohan Paul (@rohanpaul_ai):** > UBS says 60% of companies now watching AI budgets are moving to cheaper models and open-source Chinese models > The pressure is coming from extreme bills, including users spending up to $35K/month, teams exceeding quotas by 200%, and companies cutting internal AI tools from 5 to 2. > Companies are not abandoning AI, they are using model routing, which sends easy tasks to cheaper models and saves premium models for hard reasoning, code, and long-context work. > Chinese open-source models such as Qwen, DeepSeek, MiniMax, GLM, and Kimi now fit the enterprise cost curve because they can be run locally or used through cloud catalogs. > --- > news .futunn.com/en/post/75068082/ubs-group-finds-60-have-already-started-curbing-ai-spending?level=2&data_ticket=1780870170397383 > https://x.com/rohanpaul_ai/status/2070358321232839073