# AI capex — X 热门讨论 (2026-10-10 14:34 UTC)
## @Balder13946731 (Balder) · 10-10 11:49 · ♥34 ↻3 💬10 每次看媒体感叹Sam Altman对OpenAI「0持股」有多纯粹，我都觉得大家太天真了。这绝对是硅谷最高明的一场杠杆游戏。他确实没拿OpenAI的股权，但他个人的投资盘子早就过了20亿美金。玩法其实很直白：拿OpenAI的天量开支去喂饱自己的portfolio🌚 比如炼大模型需要买核电，OpenAI就去跟Helion签协议，而Sam个人持有Helion大概16.5亿美金的股份；要是缺备用芯片，OpenAI直接拍出5100万美金的采购意向给Rain AI，这公司也是他早年投的；连ChatGPT每个月收的会员费，后台都切给他有超6亿美金权益的Stripe。别人做风投，是投完钱天天求着大厂给订单。他倒好，直接捏着一个全宇宙最能烧钱的超级大客户，把百亿级别的Capex灌进自己的局。假设你在这个位子上，手里握着每年上百亿美金的采购大权，面前站着一排你自己砸了真金白银的初创企业，你会怎么批这些单子？ https://x.com/Balder13946731/status/2108887695049302486
## @OnodaCapital (Hiroo Onoda) · 10-10 13:40 · ♥33 ↻2 💬5 Argument for 60/40 is quite strong
S&P highly levered to AI. Capex slowdown will be huge benefit to bonds (spare me the ‘12% deficits cpi to 9 all assets tank’ bear porn). > 引用 @OnodaCapital: Irony is outlook for USTs could be better than in many years
- Either AI delivers material productivity, or we go into recession as capex has to cut
- Dems gaining more power and focused on higher taxes + no data centers. Prob more immigrants
- Tariffs likely down not up https://x.com/OnodaCapital/status/2108915636252664055
## @PrasoonPratham (Pratham Prasoon) · 10-09 19:19 · ♥31 ↻0 💬2 I'm no longer bearish on generative AI (mostly LLMs), but I'm still unsure about the economics of these frontier labs.
Is there enough marginal demand for their tokens to sustain trillions of $ in capex?
We need an LLM to cure cancer or prove P = NP for any of this to make sense. https://x.com/PrasoonPratham/status/2108638544491339911