# AI capex — X 热门讨论 (2026-09-13 11:07 UTC)
## @AngelicaOung (Angelica 🌐⚛️🇹🇼🇨🇳🇺🇸) · 09-13 10:08 · ♥36 ↻7 💬3 This is a most fascinating chart showing that hyperscalers went hog wild on CAPEX right after the DeepSeek moment.
According to Izabella who is a great follow (if sometimes tinfoil hattish) that cannot be a coincidence. DeepSeek showed that Chinese open source AI can deliver near-frontier performance at cabbage prices despite the chip ban. This led the US labs to go on a spending spree because they concluded scale is their last moat.
This kicked off a Highlander-esq scaling war between the US labs because the spend is unsustainable except if you are the last lab standing.
But Kimi 3 broke THAT when it showed up again, right below SOTA but with a fraction of the compute. This broke the linear CAPEX race to win logic.
And now they need to get off the capex rollercoaster and thus the focus on safety and « pacing the frontier » > 引用 @izakaminska: REMINDER: They need a credible off ramp from the hyper scaling narrative they’ve been selling for months. Safety concerns are an obvious way to justify the incoming pivot. They also justify taking the whole thing into a closed “Manhattan project” type setting at the government’s expense. This justifies the incoming lockdown.
As I predicted in August the scale narrative had run its course due to kimi. Hyper scaling wasn’t about genuine demand expectations. It was about winning the ai race by outspending and bankrupting your competitors. Kimi made that pointless.
This is no doubt the asymmetric information Bessent has been alluding to for weeks. There’s going to be a helluva lot of liquidity freed by this move. It’s all going to go flooding into USTs. Happy D day bond guys. 👇 https://x.com/AngelicaOung/status/2099077837399740500
## @imnotharsh (ImNotHarsh | 📈💸) · 09-12 21:44 · ♥32 ↻6 💬7 I have a counter to FinX calling for some bubble pop in AI stocks due to Dario's essay. The essay today is being read as a halt, but that is not really the goal here.
Dario literally says, "...pacing does not mean halting model training or technical progress, but ensuring companies take adequate time to align and safeguard their models, and for third party evaluators to confirm this."
That just means it is best to have auditors in the building and it DOES NOT change anything with the GPUs, CPUs and data centers to house them, which are still needed to run the demand that exists even years out.
Anthropic is committing to embedded third-party evaluators. Sam says OpenAI will match the evaluator step. Elon says Dario is right. None of that is a training freeze. None of that is an IPO cancellation. None of that is a capex holiday by the hyperscalers. No one is stopping spending. Let's be real.
The market mistake is treating a process change as a demand shock for silicon. It is not.
Who still buys the chips:
- Hyperscalers. AWS, Azure, GCP, Meta. Cluster buildouts are multi-year. They do not pause because a lab added desks for METR. - Anthropic. Claude still trains. Evaluators sit on the same campus as the next run. - OpenAI. Same. - xAI. Same. - Every other frontier lab that is not going to unilaterally surrender the frontier to a peer that kept training.
Safety work is compute-positive in the near term, not compute-negative. Alignment, evals, red-team, interpretability, and “take more time between capability jumps” still burn tokens and still sit on HBM, networking, and foundry capacity. A year of slower *release cadence* with the same or higher *training and inference load* is a semiconductor bull case, not a cut.
If someone wants a real demand break they need cancelled GPU orders, delayed data-center energization, or hyperscalers guiding capex down.
Today’s note does not do that. It literally says keep training, add auditors, ask Washington for a coordination waiver, keep the U.S. lead.
Read the sentence he wrote. Then look at who is still building clusters. Those are not the same trade. https://x.com/imnotharsh/status/2098890440972075047
## @totosanpartners (Phoenix) · 09-12 23:08 · ♥31 ↻1 💬0 AIの進化が速すぎて 安全対策が追いつかないことへの警告！
投資家として見るべきは これが実際にハイパースケーラーの AI向けCAPEX減額につながるのか？
ただ、中国企業がAI開発を減速させるとは考えにくい。AI覇権争いの中で、世界同時の減速は無理だよね🤔 https://t.co/Qoff38QnjA https://x.com/totosanpartners/status/2098911736128819577