# AI capex — X 热门讨论 (2026-09-17 23:11 UTC)
## @WarrenPies (Warren Pies) · 09-17 22:16 · ♥162 ↻23 💬12 Yesterday, @NickTimiraos asked Kevin Warsh what had changed over the past year to reverse his view of the economy.
Yes the Iran War has created a supply shock. But, the real change is the AI buildout. Never before has the Fed cut in the face of a capex boom until the past year. https://t.co/WA47JUCpM1 https://x.com/WarrenPies/status/2100710392368369737
## @hsu_steve (steve hsu) · 09-17 11:04 · ♥34 ↻4 💬4 AI boom or bubble? ~$trillion capex, but only ~tens of $billions in organic revenue.
Financial bubble, unless revenue doubles every year for next ~4 years.
https://t.co/xbKTPmBO8g > 引用 @momans: @hsu_steve explaining why rapid recursive self-improvement means nothing for AI valuations until real-economy businesses -- not VC-backed startups -- start organically spending trillions on the tech. https://t.co/LET3zJH3bW https://x.com/hsu_steve/status/2100541418611573210
## @RosannaInvests (Rosanna Prestia, MBA) · 09-17 21:57 · ♥32 ↻4 💬5 $IREN delivered about 3 megawatts of AI compute a year ago. This year it's 480, and they're guiding 1.2 gigawatts for 2027.
The miners spent five years buying power nobody else wanted. It's now the scarcest input in tech.
That's the whole neocloud trade.
The knock on neoclouds is always "who pays for the GPUs."
At $IREN the customers do. Prepayments cover close to half the GPU capex, ARR guide is $4B, and the 2GW Sweetwater hub tied into ERCOT on Sept 8.
JPM upgraded two notches to $65 this week.
The power contracts were the asset. ✅ https://x.com/RosannaInvests/status/2100705851342528958
## @brxckinridge (breck) · 09-17 20:33 · ♥32 ↻3 💬4 From Wafers to Watts to Basis Points https://x.com/brxckinridge/status/2100684666693197934
## @IEObserve (Intl Econ Observe) · 09-17 10:14 · ♥30 ↻1 💬1 AI泡沫論的人眼要夠瞎，才會看不到終端用戶付錢的指數性成長，這還是在AI還沒那麼好用的情形之下
不過 Altimeter的Gestner也有提出現在市場情緒關注的checkpoints 是什麼
https://t.co/Ucd8pwf5W2
———
GPU 使用量、token 數量與算力需求都可以很高，但最後仍然需要有人付錢，才能讓這些資料中心長期獲得合理報酬。
過去一家軟體公司如果能在四、五年內做到 10 億美元營收，已經屬於極少數成功案例；現在市場討論的，卻是領先 AI Lab 能否在一年內增加數百億美元營收。
這也是為什麼 Gerstner 認為，接下來最重要的市場數字可能不再只是 NVIDIA 又賣了多少 GPU，而是 OpenAI 與 Anthropic 每個月到底產生多少營收。
因為這個數字才真正決定目前龐大的 AI CapEx 能不能被消化。 https://x.com/IEObserve/status/2100528906977890491