@fchollet：François Chollet 拆解 AI 泡沫的五层定义：技术可行≠非泡沫，繁荣本身亦不能证伪泡沫，关键在于基本价值…
Source: https://x.com/fchollet/status/2064740102463725853
Author: @fchollet (fchollet)
Original Date: Wed Jun 10 16:02:56 +0000 2026
Likes: 1849 | Retweets: 215
Quality Score: 5
Fetched: 2026-06-22 20:24:54
English Original
**Some considerations that many folks seem not to get:
1. It can be a bubble even if the tech works. (For instance, if the tech doesn't have a high-demand use case.)
2. It can be a bubble even if the tech works and has strong product-market fit. (For instance, if the tech cannot be economically viable.)
3. It can be a bubble even if the tech works, has strong product-market fit, and has a path to eventual economic viability. (For instance, if profitability takes too long to achieve or makes margin/competition assumptions that fail to materialize.)
4. It can be a bubble even if the tech works, has strong product-market fit, and is currently highly profitable. (For instance, if demand has a hard ceiling and growth stops once the ceiling is reached.)
5. It can be a bubble even if the tech works, has strong product-market fit, is currently highly profitable, and has unlimited future demand.
Literally all it takes for something to be a bubble is for lots of people to over-enthusiastically bet their money on it, and subsequently get panicky.
Importantly, bubbles can be attached both to things that are completely hogwash, like the Metaverse, and to world-changing developments like the Internet or railways. Bubbles don't care. They're brought into existence by the thoughts and feelings of investors, not by actual tech or products.
"The bubble has burst" doesn't mean "the tech didn't work" or "people stopped using the tech." It only means that people got panicky, investor money dried up, and valuations collapsed. Internet adoption didn't stop in 2000.**
Tweet Details:
- Author: fchollet
- Bio: Co-founder @ndea. Co-founder @arcprize. Creator of Keras and ARC-AGI. Author of 'Deep Learning with Python'.
- Likes: 1849
- Retweets: 215
- Created: Wed Jun 10 16:02:56 +0000 2026
- URL: https://x.com/fchollet/status/2064740102463725853
Thread / Replies Context
@Candlejack34981 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet If a transformative technology is actually transforming people, people will not need to be told as they will be busy transforming.
*link*
@luyun0120 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet 产品价值不是发布会上的PPT决定的，是用户用完还回不回决定的。 「许多人似乎没有明白的一些考虑因素：」
*link*
@Rahu159 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet By this definition @SpaceX IPO falls into this as well.
*link*
@luyun0120 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet This is the clearest articulation of why 'it works' isn't enough. The third consideration often gets overlooked: even valid tech with PMF can be a bubble if the economics don't work out at scale. The AI industry is about to learn this the hard way.
*link*
@tarunyaa_ (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet In our case, the question is whether application-layer demand materializes before the buildout completes in 2027-28.
*link*
@behindceo (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet people get caught up in hype instead of focusing on real problems the tech solves just because something works doesnt mean it will sustain value long term
*link*
@IntuitMachine (👍 3 · 🔁 1) *In reply to 2064740102463725853*
@fchollet Over supply creates bubbles. This doesn't mean the technology does not work. It only means that the demand did not catch up to the supply.
*link*
@MarcL_Guru (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Curious how many French renewable projects sit between your points 2 and 3. Complément de rémunération guarantees revenue, but RTE grid queues make delivery timelines impossible. Awarded vs operational MWs tells the story.
*link*
@tusharpatel0 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet legit
*link*
@baclace (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet There absolutely is a bubble as seen in the gigantic expansion in data centers. But wait, the the retail investors will arrive shortly. When the cashier at the grocery store tells me about the money they made on their AI investments, then I know the end of the boom is near.
*link*
@gunjansiddha (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet The giants of the internet boom during late 90s and early 2000s went down quick and bad for a long term. They all surpassed their peak valuations again recently - IBM, Intel, Cisco and such.
*link*
@woo0057 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet so... calls on spacex and puts on spy?
*link*
@woo0057 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet nice
*link*
@shiyam_kashfiq (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Exactly-valuation swings follow hype, not product merit
*link*
@ronitkd (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet The dot-com point is the one people skip. The crash killed valuations, not the tech. The people who kept building through it owned the next decade. Panic clears out the tourists. The shippers stay.
*link*
@jiejur (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet And, conversely, "numbers go up" does not automatically means "it's a bubble !"
*link*
@MuzammilData (👍 1 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Bubbles aren’t about whether the tech works they’re about whether future expectations embedded in prices keep attracting new buyers. Once that breaks, valuations collapse even if adoption and profits remain intact.
*link*
@PlastiqSoldier (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Sure, but it's not a bubble.
*link*
@LuoYaoshen (👍 1 · 🔁 0) *In reply to 2064740102463725853*
Exactly. The logic breaks down at the ROI. Code is a means to an end, not the product-market fit itself. A flawless AI-coded software that costs $1000 to infer but only generates $100 in revenue is a textbook bubble. As marginal utility plateaus, the market will experience a brutal regression to economic common sense.
*link*
@moyingy66 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet agree
*link*
@wonderwomancode (👍 2 · 🔁 0) *In reply to 2064740102463725853*
@fchollet I graduated while the dotcom bubble burst was still shaking out. Watched over 600 layoffs happen in my building at WDIG while I was doing my internship. I still went on to carve out a niche consulting companies with migrating to the cloud and making their systems online.
*link*
@broadfield_dev (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet TL;DR discount GPUs and RAM soon.
*link*
@gmtreedev (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Francois waiting for the bubble to burst so we can finally have investments into real efficient AI
*link*
@waral555 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Very insightful and not at all useful unless you can predict when to profitably exit in a way that doesn't result in you buying back in higher.
*link*
@MLCatttt (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet the missing point 4 is that the capability curve and the capital curve are decoupled. the tech can keep compounding straight through the financing unwinding, those are two different clocks
*link*
@ArcanesValor (👍 1 · 🔁 0) *In reply to 2064740102463725853*
@fchollet @provisionalidea I would argue that the further you are along this list the more likely a bubble is to form and the larger the bubble is likely to be.
*link*
@shzhv13 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Bubbles are a capital allocation bug, not a tech verdict. The signal is unit economics under load. Cost per useful token vs value created.
*link*
@mfarles (👍 1 · 🔁 1) *In reply to 2064740102463725853*
@fchollet Before all of that, 0. They knew they have to little costing too much but still wanted to rule the world. Too dumb to create the real thing, too dumb and greedy not to go too far into scam land.
*link*
@breenemachine (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Are you using AI to code now, Francois?
*link*
@chaudstopper (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Tulips have all these characteristics.
*link*
@SereneInvesting (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Very nice post!
*link*
@AIBlancos (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet chollet is right about the bubble mechanics, but the part people miss is that post-bubble consolidation usually picks *winners*, not kills the technology. AWS, google, modern ai infra all came out of the 2000 wreckage.
*link*
@assumptology (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet To express it succinctly: Separate the money from the tech and the short from the long term.
*link*
@permataste (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet If few generations of models saw no qualitative improvement and wall idea started seriously floating around, this would seriously pop it off. Starting with nvidia. But ofc everyone is feeling the agi so very unlikely
*link*
@lordofblocks (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet The internet didn’t stop being the internet after 2000. Valuations still collapsed. Both things happened.
*link*
@sudo_trader (👍 1 · 🔁 0) *In reply to 2064740102463725853*
@fchollet There's always a bubble if you look hard enough, and you'll lose more money trying to prepare for the bubble pop than you would if you just ignored all the noise.
*link*
@InverseMarcus (👍 1 · 🔁 0) *In reply to 2064740102463725853*
@fchollet OK but its not a bubble
*link*
@vannibombonato (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet It’s not AGI and it’s nowhere close, but average intelligence on demand at scale across everything is an immense market.
*link*
@BugN07 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
I agree that bubbles are ultimately driven by investor behavior.
However, investor behavior is often downstream of narrative.
Narratives determine what people believe, what risks they fear, and what futures they invest in.
A bubble is not only a financial phenomenon.
Sometimes it is also a narrative phenomenon. *link*
@mathepi (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet all the crap about a 'slowdown' or a 'pause', now that the world is in hock to data center construction, but before the promise is fulfilled...is the greatest rug pull in history.
Sue Dario's ass next time he yaps about a pause! *link*
@RexExLux (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet bubbles are necessary in big tech cause of pharaonic investment needed to create a new economy. the bubble is already bursting, it started when companies are starting to restrict the best models out there
*link*
@gentile_captial (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet ur a fintwit influencer now, eh? $sive
*link*
@angelaharris0 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Yes🙌
*link*
@PeterRex (👍 2 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Yep. Bubble is about investor behavior, not the tech itself. Internet got the same treatment in 99, pop didnt stop the transformation
*link*
@AutoCrossBow0 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Mostly agree except for the fact that Metaverse is not hogwash just early.
Closer to hogwash I would say is consumer sentiment of lithium battery powered electric vehicles.
Ai is not a bubble just hard for people to grasp and appreciate what they can't physically see. *link*
@aitization (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Well said.
*link*
@ArtCfartsy_0 (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet The same company that won the last bubble are in the race for AGI as we speak!
*link*
@MarcusR_67 (👍 7 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Exactly this. You could argue that the bursting of the 2000 tech bubble was a net positive for the progress of the Internet. It certainly did nothing to slow it down, and the bursting of the bubble reallocated capital from frivolous projects into more productive uses.
*link*
@FredPicks (👍 0 · 🔁 0) *In reply to 2064740102463725853*
@fchollet Sure. But it can also not be a bubble, even if stock prices have an insane run
*link*
中文翻译
François Chollet（@fchollet）提出关于"AI 泡沫"的五层判断框架：
1. 即使技术可行，也可能是泡沫——比如该技术缺乏高需求使用场景。 2. 即使采用率持续上升，也可能是泡沫——比如估值远超基本面的合理水平。 3. 即使能盈利，也可能是泡沫——比如企业利润建立在终将崩盘的债务或补贴之上。 4. 即使行业里多数公司都在赚钱，也可能是泡沫——比如少数赢家的盈利被大量亏损公司稀释。 5. 即使当前处于繁荣周期，也可能是泡沫——只要其中存在不可持续的要素。
Chollet 强调：判断泡沫应聚焦"基础价值是否被可持续的现金流支撑"，而不是表面繁荣与否。
思考：技术繁荣 ≠ 商业可持续。当前的 AI 投资热度是否建立在了可长期维系的现金流与真实需求上，值得更冷静的审视。
关键要点 / Key Takeaways
- 核心观点： François Chollet 拆解 AI 泡沫的五层定义：技术可行≠非泡沫，繁荣本身亦不能证伪泡沫，关键在于基本价值是否被可持续现金流支撑。
- 作者立场： fchollet（@fchollet）在 AI 行业具有较高影响力，长期关注 AGI、可解释性与开源生态。
- 行业意义： 推文讨论的议题与当前 AI 产业演进方向高度相关。
*Source: https://x.com/fchollet/status/2064740102463725853* *Translated by openclaw pipeline · 2026-06-22 20:24:54*