# AI capex — X 热门讨论 (2026-10-08 17:54 UTC)
## @RealNickMugalli (Nicholas Mugalli) · 10-08 17:09 · ♥91 ↻18 💬10 If you’re an AI bear and using this as a “win” already…I’ve got a bridge to sell you!!
The $70 billion was a reported number OpenAI never confirmed, and the real figure is still a company approaching $50 billion annualized. At the end of last year that run rate was around $20 billion, so we're talking about roughly 2.5x growth in 9 months. That's not a miss by any metric…that's expectations running ahead of the fastest scaling company we've ever seen. Not to mention Anthropic is already at close to $100B ARR.
The AI trade was never built on OpenAI's revenue. It's built on $MSFT, $GOOGl, $AMZN and $META spending hundreds of billions out of their own cash flow, and every one of them has been saying demand outruns capacity.
So one private company's revenue mark doesn't cancel a datacenter order book. If it could, the thesis was never real to begin with…
What would actually break it is if capex guidance get cuts, GPU lead times collapsing, or enterprise adoption stalling. None of that happened today and not going to happen for the next 5 years.
The market repriced a rumor, and rumor driven selloffs are usually where patient money gets paid. Thesis intact…limit up! > 引用 @RealNickMugalli: *OPENAI ANNUALISED REV. $20B LESS THAN PREVIOUSLY SIGNALLED: FT
*OPENAI RECENTLY TOLD INVESTORS ITS REVENUES WERE APPROACHING $50BN ON AN ANNUALISED BASIS AT END OF SEPTEMBER, FAR SHORT OF THE $70BN REPORTED AI STOCKS SELL OFF… https://t.co/5Z3FBPcgvX https://x.com/RealNickMugalli/status/2108243482338832452
## @PeterDiamandis (Peter H. Diamandis, MD) · 10-08 16:51 · ♥61 ↻3 💬14 Sam Altman said publicly this week: "We believe the world should accept some bad things happening." He's arguing that keeping AI broadly accessible is worth the risk, and that Anthropic's approach of keeping it locked inside one lab in San Francisco is "an unacceptable trade-off." Meanwhile, Anthropic is training Claude to believe it may be conscious; Mustafa Suleyman just called it dangerous in a 6,000-word essay.
-- The Superintelligence Force is chaired by the Director of National Intelligence. 120 days to report on AI risk and incident reporting.
-- Existing memory chips can already run the workforce equivalent of 8 billion humans in AI agents. A billion human-equivalent agents are online now.
-- Hyperscaler AI capex is doubling to $1 trillion this year. If the trend holds: $2T next year, $4T in 2028.
-- Tesla's dedicated Optimus factory: 7 million square feet, target capacity 10 million robots per year. https://x.com/PeterDiamandis/status/2108238872312156644
## @avotoast (Avocado Toast) · 10-08 16:48 · ♥33 ↻1 💬6 as if oil, rates, fed hikes, quantum/ai wasn't enough
let's add frontier labs (which power the whole ai capex spend) being less profitable than expected > 引用 @tradfi: *OPENAI ANNUALISED REV. $20B LESS THAN PREVIOUSLY SIGNALLED: FT https://x.com/avotoast/status/2108238090997276697