# AI capex — X 热门讨论 (2026-09-29 09:18 UTC)
## @a16z (a16z) · 09-28 17:08 · ♥120 ↻8 💬20 TypeSafe AI's Diogo Almeida expects an inverse SaaSpocalypse, and Martin Casado calls it "SaaSpalooza":
Ben: "When the coding agents came out, it was the SaaSpocalypse, and all their values dropped through the floor. Then when Jev came out, every SaaS company is like, 'This is the greatest thing ever.'"
Diogo: "In the SaaSpocalypse story, the story I feel has panned out really poorly is that software is very cheap and perhaps easy to replicate. I could believe the former. I could not believe the latter, because a lot of the stuff happens beneath the hood."
"I think SaaS will be one of the largest winners of the whole AI game. I want to work really, really well with the biggest, most boring... SaaS companies because I think they're the best positioned to know what workflows to automate, what people need. That's their bread and butter."
"Software is always a CapEx investment. You spend it ahead of time to make the experience better, and that gets distributed to all that mass of users."
"I'm not going to forecast anything about the financial markets. But as far as capabilities go, I think it's going to be like an inverse SaaSpocalypse. I'm so jazzed about it. I should make a name..."
Martin: "SaaSpalooza."
@CompleteSkeptic @bhorowitz @martin_casado > 引用 @a16z: TypeSafe AI's Diogo Almeida with a16z's Ben Horowitz and Martin Casado on Jev, the model built to live inside software:
Diogo's elevator pitch for Jev is a simple question - where is all the automation?
AI is unbelievably smart, but outside of chatbots and coding agents, it hardly touches any real work. His diagnosis is the industry built models that generate text for humans to read, and software can't consume that output.
Jev reads natural language and returns a choice from a set of options with a confidence level assigned to each, so developers can build programs that reason about intent and make probabilistic decisions rather than relying on human interpretation.
TypeSafe's philosophy is "We build prod, not God."
0:50 "Where the f**k is all the automation?" 2:50 Jev vs. Claude Code and Codex 6:55 Jev is a classifier and classifiers are sick 7:40 Chat vs. code: is Jev a slider? 9:00 Diogo: From mathlete to Kaggle to OpenAI 12:20 "We build prod, not God" 15:55 Reliability over demos 16:55 2021 thoughts: RLHF is AGI? 20:45 Optimizing for the wrong use case 21:50 Is the real world too messy to automate? 25:00 Nobody expected the Jev launch 26:35 Three kinds of reliability 28:05 Good at syntax, bad at architecture 30:00 The inverse SaaSpocalypse 33:40 Why coding agents automate so little 36:05 Probabilistic programming returns 38:45 Jev as the UDP-to-TCP layer for AI 40:20 The 5 stages of grief for embedding AI 41:30 Utopia: AI that actually does what you mean
YouTube: https://t.co/A3gcHY3iLa
@CompleteSkeptic @typesafeai @bhorowitz @martin_casado https://x.com/a16z/status/2104619274254114848
## @fednmad (fednmad) · 09-29 06:38 · ♥41 ↻8 💬1 베센트, 월가의 비주류 전략가 데이비드 저보스를 재무부 고문으로 영입 https://x.com/fednmad/status/2104823025854300219
## @QuintenFrancois (Quinten) · 09-29 06:47 · ♥41 ↻3 💬3 This dropped less than a week after I wrote this thread.
Anthropic did $4.6B in revenue in 2025.
Operating loss: $8B Future cloud + compute commitments: $518B
AI is revolutionary… but the economics behind this capex race are still absolutely insane. > 引用 @QuintenFrancois: 1/ I genuinely think AI will change the world.
I’m far less convinced investors will make the money they expect.
The more I look at the financing, the more I worry about who gets the bill if this goes wrong.
And the more interesting Bitcoin becomes to me. 🧵 https://t.co/UHVrFWxj4F https://x.com/QuintenFrancois/status/2104825432776626322
## @WonhaengL (닥터음양) · 09-29 08:57 · ♥38 ↻3 💬0 조정이 일어날지라도 왜 음봉에 사라고, 롱을 부르는가.
현재의 장기채금리상승이 인플레이션이 문제가 아닌, AI 및 그 인프라가 동시에 자본을 빨아들이는 수요 충격으로 보면 이해가 쉽습니다. 이러한 여건에서 Fed의 금리인상은 장기채 금리를 눌러주는 도구가 되기 어려울 수 있다고 봅니다.
기존 모델에서 금융여건이 타이트해져 성장·투자·인플레 기대가 꺾이고, 수요가 억제되는 것이 가시적으로 들어온 그 다음에야 장기금리가 내려오는데, 금리가 이미 높은데도 빅테크, 데이터센터, 전력망 투자가 멈추지 않는다면 Fed가 단기금리를 더 올려도 자본수요 자체가 안 죽어서 장기금리가 같이 내려오지 않습니다. 오히려 “성장이 강하니까 더 올려야 한다”는 기대를 키워 장기채 금리를 밀어올릴 수 있습니다.
빅테크는 예상 수익률이 높아서 올인 금리(국채 + 스프레드)가 좀 올라도 찍습니다. “비싸도 찍는다”는 일이 벌어지는 중입니다. 당장 투자를 위한 실탄이 필요하니까요. 지금 AI투자는 전쟁 국면과도 같습니다. 대신 시장은 그 비용을 회사채 스프레드만으로 다 받지 않고, 무위험금리(국채) 자체를 올려 전체 자본가격을 재조정합니다. 국채가 “안 팔린다”기보다, 같은 현금을 놓고 더 높은 실질수익률을 요구하는 겁니다.
금리가 상승하면 문제가 되는 것은 빅테크가 아니라 주택, 한계기업, 비AI capex, 만기가 긴 프로젝트파이낸스입니다. 자본배분이 AI로 쏠리면 국채도, 일반 회사채도 같이 비싸집니다. 즉 Fed가 기준금리를 올리면 회사채 금리의 레벨은 분명히 올라갑니다. 그런데 그 상승이 빅테크의 회사채 발행을 멈추게 하진 않고, 국채금리 + 스프레드를 동시에 끌어올린 채 발행이 계속될 가능성이 큽니다. 그러면 Fed 인상은 장기국채 상승을 제약하기는커녕, 단기금리를 올려 커브 전체를 더 비싸게 만들 수 있습니다.
정리해 보자면, 지금 장기채 금리를 끌어올리는 동력은 “정책금리가 낮다”가 아니라 “써야 할 돈이 많다”에 가깝습니다. 기준금리를 올리면 단기와 회사채 레벨은 올라가고, 투자를 미룰 수 없는 발행자는 더 비싼 회사채를 찍습니다. 그 결과 국채 수요가 상대적으로 약해 보이거나, 최소한 국채도 더 높은 실질금리를 줘야 하는 상태가 유지됩니다.
그리고 장기채금리를 누를 수 있는 궁극적인 해법은 다음일겁니다. 1. AI 회수기간·전력비용·가동률이 의심받기 시작할 때: 이미 메타의 muse 및 AI 속도조절론이 나오기 시작했죠. 저는 이 것이 금리 상단을 누르기 위한 논리 공급 용도도 있다고 생각합니다. 2. 해외자본·저축 공급이 늘거나, 생산성 상승이 금리에 선반영될 때: 이건 아직 가시적으로 나타나기엔 이릅니다. 3. 재무부가 장기물 공급을 줄일 때: 이게 가장 중요합니다. 이건 미국-이란 전쟁의 진행추이와 직접적인 관계가 있죠. 앞으로 이란과의 협상 혹은 전쟁이 어떻게 진행될지 모르지만, 그래도 장기채금리 상단을 무한정 높일 수는 없는 노릇이기 때문에 어느 시점에서는 관리 들어갈 것으로 생각하고 있습니다.
결국, 장기물 공급과 관련된 재정의 문제이죠. 앞으로 이 논리가 심화된다면 -어쩌면 향후 시장이 예측하는 금리인상 전망이 높더라도- 의외로 향후 FOMC에서는 금리 동결안이 나올 수 있습니다.
그리고 흔하 말하는 잡주, 즉 중소형주도 현 시즌을 지나가면 옥석이 가려질 것으로 생각합니다. 생존을 위한 현금을 지니고 있는 기업, 그리고 성장이 분명히 나타나는 기업에 시장은 가중치를 부여할겁니다. 저는 그 시금석, 혹은 탄광 속의 카나리아가 ionq라고 보고 주시하고 있는 중입니다. https://x.com/WonhaengL/status/2104858209374892481
## @Namzes_G (Namzes Cycles) · 09-29 03:12 · ♥33 ↻2 💬5 Historically good economy was linked to high approval ratings for U.S. presidents; with the stock market being the main transmission mechanism / signal for the economy. Problem is that it has become the focus of the administrations at the expense of the inflation which is a lot more important for average Americans who don’t own 7-8 fig portfolios. Biden was guilty of it and Dems lost by a landslide, now Trump is in identical position and is obsessed with the stock market while trying to gaslight Americans with cooked inflation numbers.
The problem of course is that the only way to slow down inflation is to reduce the deficit and tighten monetary policy via reduced money supply (QT + slower bank loans / ai capex growth via higher rates). The short term pain would be in the stock market and would also reduce consumption in K shape economy so obviously no one wants to see the stock market down even 5% for more than a few days.
Since no one wants to stop running war time deficit at the peak of economic cycle, you have to let market discipline work its way and watch politicians come in and out until they figure out that you can’t eat stocks for dinner. > 引用 @Namzes_G: By the time Powell and co will realize poors can’t eat Nasdaq for dinner he’ll be out of job Trump will not be any better (probably worse) but economy is on the president always - so biden will be gone https://x.com/Namzes_G/status/2104771358890614971
## @MentoviaX (MentoviaX) · 09-29 04:46 · ♥35 ↻1 💬2 $MU Micron reports Wednesday after market close. Here are earnings preview and trading scenarios:
Street sits at $31.30 EPS on $50.6B revenue, essentially the company’s guide of $31.00 ±$1.00 on $50.0B ±$1.0B with ~86% gross margin. Options price an 11% move. Michael Burry now in panic mode replacing his huge short positions with less sizable put positions.
Positioning context: an 11% implied move on this revenue base tells you the options market agrees this is binary. Put skew has been bid since Burry’s disclosure, with open interest stacked at the 10% out-of-the-money strikes expiring Friday.
Crowded short plus beat-and-raise is a squeeze; crowded long plus tempered guide is a rug pull. Know which crowd you are in before Wednesday close.
THREE scenarios:
1. “Beat and Raise” — 60% probability.
MU prints $34.20 EPS on $53.1B revenue with gross margin expanding to 87.5%. DRAM contract pricing up 8% sequentially, HBM4 qualification ramping ahead of plan, and data-center revenue now 55% of the mix. Management guides FQ1 to $34–$36 EPS and raises FY27 capex to $22B — a demand signal, not a supply warning.
Stock: +12% to +15%. The short thesis dies on the print, with shorty covering accelerates the move.
Key tell: any quantified AI-accelerator memory backlog extending into 2027.
2. “Beat, Then Tempered Guide” — 30% probability.
MU clears the bar modestly: $32.50 EPS on $51.4B revenue, margins flat at 86%. The beat comes from NAND recovery, not DRAM acceleration. But FQ1 guidance lands at $30–$32, under the $33 whisper, with management citing “normal seasonal inventory digestion” among smartphone OEMs, or supply chain bottleneck.
Stock: flat to -5% initially, with direction set by the call’s tone on December-quarter DRAM contract pricing. This is the consensus trap — good quarter, bad narrative.
3. “The Cycle Turns” — 10% probability.
MU misses: $28-29 EPS on $48.6B revenue, gross margin slipping to 83%. The culprit is Chinese DRAM capacity (CXMT) flooding legacy nodes, dragging blended ASPs down 6%, while enterprise SSD demand pauses ahead of platform transitions. FQ1 guide gets cut to $26–$28.
Stock: -10% to -12%, and Burry’s short bet becomes the cover story. The structural bull case — HBM, the new 512GB DDR5 module — survives intact, but the multiple compresses as the market reprices 2027 EPS from $120 down to $95.
What I will be watching on the call:
1. DRAM contract pricing commentary for the December quarter. 2. HBM revenue run-rate disclosure. 3. Inventory days — anything above 140 is the bears' smoking gun. 4. Capex discipline versus capacity additions.
Base case is Scenario 2. The quarter clears the bar; the guide sets the stock. But I honestly feel Scenario 1 is more likely to happen.
For traders, using ETF maybe a better way to balance between risks and rewards.
(BIASED PERSONAL OPINION. NOT INVESTMENT ADVICE. I AM BULLISH AND LONG ON MICRON.) https://x.com/MentoviaX/status/2104794984880676892