Learning Library

← Back to Library

Beyond the AI Bubble Hype

Key Points

  • A growing “AI bubble” narrative has emerged, fueled by the disappointment around the botched GPT‑5 rollout, high‑profile layoffs in Meta’s AI division, Sam Altman’s own admission of a bubble, and an MIT study highlighting the high failure rate of enterprise AI projects.
  • The hype‑to‑doom swing is partly driven by a collective need for a dramatic story, as the initial excitement over GPT‑5 quickly turned into a counter‑reaction seeking a new narrative.
  • The MIT research underscores that successful AI adoption requires strong leadership, cultural change, and clear high‑value use cases—factors many organizations are still lacking.
  • Despite the negative sentiment, the chatbot market is reaching saturation, meaning incremental gains from larger models are becoming less perceptible to end users.
  • A more balanced view recognizes that while certain AI applications face diminishing returns, the broader AI landscape still holds realistic opportunities beyond hype‑driven hype.

Full Transcript

# Beyond the AI Bubble Hype **Source:** [https://www.youtube.com/watch?v=Sno3eqzgmtA](https://www.youtube.com/watch?v=Sno3eqzgmtA) **Duration:** 00:11:57 ## Summary - A growing “AI bubble” narrative has emerged, fueled by the disappointment around the botched GPT‑5 rollout, high‑profile layoffs in Meta’s AI division, Sam Altman’s own admission of a bubble, and an MIT study highlighting the high failure rate of enterprise AI projects. - The hype‑to‑doom swing is partly driven by a collective need for a dramatic story, as the initial excitement over GPT‑5 quickly turned into a counter‑reaction seeking a new narrative. - The MIT research underscores that successful AI adoption requires strong leadership, cultural change, and clear high‑value use cases—factors many organizations are still lacking. - Despite the negative sentiment, the chatbot market is reaching saturation, meaning incremental gains from larger models are becoming less perceptible to end users. - A more balanced view recognizes that while certain AI applications face diminishing returns, the broader AI landscape still holds realistic opportunities beyond hype‑driven hype. ## Sections - [00:00:00](https://www.youtube.com/watch?v=Sno3eqzgmtA&t=0s) **Debunking the AI Bubble Narrative** - The speaker identifies four drivers—storytelling cycles, Meta layoffs, a botched GPT‑5 rollout, and an MIT study—behind the current “AI bubble” panic and argues for a more grounded perspective on AI’s state. - [00:04:08](https://www.youtube.com/watch?v=Sno3eqzgmtA&t=248s) **Exponential AI Progress & Chip Scarcity** - The speaker asserts that AI performance is improving exponentially across unsaturated benchmarks like MER, while a shortage of compute chips limits model upgrades, indicating massive demand that is often misunderstood in studies such as MIT’s. - [00:07:31](https://www.youtube.com/watch?v=Sno3eqzgmtA&t=451s) **AI's Exponential Power-Law Returns** - The speaker argues that AI delivers outsized, exponential gains—following a power‑law distribution—making it an existential bet for firms and prompting massive, rational investment despite the high risk of failure. - [00:11:50](https://www.youtube.com/watch?v=Sno3eqzgmtA&t=710s) **Optimistic View on AI Future** - The speaker expresses curiosity, dismisses fears of an “AI winter,” and hopes to shift the prevailing narrative about artificial intelligence. ## Full Transcript
0:00A number of things are coming together 0:01to drive a narrative that we're in an AI 0:03bubble today. I saw a conversation just 0:06on X last night basically saying the 0:08death of AI is near, right? Like we have 0:10the profits of doom out. I want to lay 0:12out for you why I think that's 0:14happening. And then I want to lay out 0:15for you the pieces we're not looking at 0:17as a community as we talk about AI. And 0:20last but not least, I want to put 0:21together a story that I think is more 0:23accurate about where we're actually at 0:25in AI at the moment. Less hype, more 0:28reality. That's what I do. So here are 0:30the four things that I think are driving 0:31this are we in a bubble death of AI 0:33narrative. Number one, people need a 0:35story. There was a massive story and 0:37swing around GPT5 hype. It was kind of a 0:40botched roll out and people need a 0:42counter reaction. People need to come 0:44back and have a different take now. And 0:46so I think the need for narrative swing 0:48and narrative drama is part of the 0:51challenge here. Number two, reports of 0:53layoffs at Meta. So the AI division at 0:55Meta was widely reported to be 0:57restructuring. there have been cut cut 0:59offs and challenges. Well, that's a part 1:02of the layoff narrative, right? That's a 1:04part of the AI in trouble narrative. 1:06Number three, Sam Alman himself admitted 1:08the GPT5 rollout was botched and 1:10infamously said, "Yes, we're in an AI 1:13bubble or there's some elements of an AI 1:14bubble in what we're doing." And then 1:16number four, around the same time all 1:18this was happening, an MIT study came 1:20out saying that most enterprise AI 1:22projects fail, which is not new. It's 1:24it's yet another study showing that this 1:25is a high-risisk highreward kind of 1:27activity and that organizations really 1:29struggle to get it right at the team and 1:30above layer even as we see individual 1:33productivity gains. Ironically, a lot of 1:35the things that I've been emphasizing 1:37are things the MIT study called out 1:39like, hey, you need to have the right 1:41leadership, you need to have a culture 1:42change moment, you need to define a high 1:44value use case. I could go on, etc., 1:46etc. But all these four things came 1:48together, right? People saw layoffs. 1:50They saw that they needed a narrative 1:51after GPT5 was disappointing. They saw 1:53Sam saying the word bubble. They saw 1:55this thing on enterprise AI studies 1:57failing and it was like, you know what? 2:00That's it. That's it. We're done. We're 2:02in a bubble. It's over. And so people 2:04just kind of swung the pendulum swung 2:06back and the narrative has exploded from 2:07there. So I want to suggest to you that 2:10a more correct take includes the 2:11following five elements or following 2:14five facts that we're not really paying 2:16attention to. Number one, the chatbot 2:18use case is indeed getting saturated. 2:20This was reported by Sam in like an 2:23interview right after the one where he 2:24talked about the bubble. In other words, 2:26if you're in the chatbot, you're not 2:28necessarily going to see tons of 2:31tremendous gains anymore, no matter how 2:33smart the model gets because people 2:35don't necessarily perceive the progress 2:36in the chatbot because the AI is about 2:38as good in the chatbot as it's going to 2:39get. So, famously, what Sam said in that 2:42conversation with Chad, GPT6 is coming 2:44and memory is going to get better, but 2:46really the chat use case is kind of 2:47saturated. I think he's right. I don't 2:49think we have a lot more to gain from 2:51the chat use case. Number two, I think 2:53we're forgetting that progress is moving 2:54to agentic and complicated use case, 2:56which is sort of a correlary to the 2:58chatbot, right? And those use cases are 3:01hard for people to understand. I'll give 3:02you an example. There was a big 3:04conversation on X over the last couple 3:07days around whether GPT5 Pro did new 3:10mathematics when it was assigned a new 3:12theorem and did a new proof for 3:13something that a human hadn't done. And 3:15the consensus seems to be it was new. It 3:18was correct. It is a milestone but it is 3:20a different kind of innovation than we 3:23get from a human. Humans are good at 3:25creativity, intuition and the models 3:28that we have today are good at brute 3:30forcing innovation forward. And so it 3:32was in a position where it could brute 3:34force a series of calculations around a 3:36defined problem space and get to a new 3:37proof that hadn't been done before. And 3:39it did it. And that lines up with what 3:41we see in other innovation stories where 3:44we see that these models are very very 3:46good at certain kinds of innovation that 3:48really do push the field forward but 3:50they aren't doing the same work as 3:52humans and that nuance often gets lost 3:54and that's a great example of how 3:55complex agentic use case analysis is 3:58getting and assessment is getting. I 3:59don't know the math either. It's hard 4:01for people to understand or experience 4:03where the progress is. Fact number three 4:04that I think is getting forgotten. 4:06Progress is demonstrabably continuing at 4:08exponential rates. We have any benchmark 4:11that is not saturated is showing 4:12continued strong gains. I think my 4:14favorite is MER right now because it 4:17just doesn't have a top. All it does is 4:18it measures how long a task takes a good 4:20human and then it says can an AI do it 4:2350% of the time. Now I'm the first one 4:25to say 50% is a low bar, but at least 4:27it's a consistent bar. And we keep 4:29showing exponential gains on that as a 4:31use cases get stronger and we're not 4:33bottoming out. That's not slowing down. 4:35We keep doubling every few months. 4:36Number fourth, we are still 4:38underallocated on chips. In the same 4:40interviews that got blown up around the 4:42world around AI and bubbles, Sam 4:44admitted he could release a smarter 4:46model, but he lacks the chips to do it. 4:48Anthropic is also famously 4:50underallocated on chips. Everyone's 4:52using them for coding, and they just 4:54can't get enough chips. They can't do 4:56it. So, in that world, if they're 4:58underallocated on chips, it means they 5:00sense tremendous demand, which is backed 5:03up by what we see from the MIT study. If 5:0595% of orgs are failing at AI, that's 5:0895% of 100% who are desperately trying 5:11to get into AI, that's the demand. 5:13Ironically, the MIT study was read as 5:16reinforcing uselessness when what it 5:17should have been read at is reinforcing 5:20the insane 5:22cost benefit that organizations are 5:23running to get AI correct. Like they are 5:26doing absolutely anything they can to 5:28force their way in the door. Fact number 5:30five, teams are refocusing now that the 5:32path to the next leg of gains is mapped 5:33out. I have been in a lot of corporate 5:36restructurings. It's very typical once 5:37you bring in fancy new talent like Meta 5:40has to restructure and that is exactly 5:42what they did. And the path to the next 5:44leg of games is around inference and 5:46Meta has grabbed a bunch of people who 5:48are good at the next leg of AI computing 5:51and they're just refocusing to do that 5:52well. I don't think that's that big a 5:54surprise frankly, but it got fed into 5:56the story. So if you put this all 5:57together, you get a story of continued 5:59progress on high-v value use cases. 6:01Continued demand for chips. Ironically, 6:03continued demand for intelligence backed 6:05up by everybody saying they don't have 6:07enough chips to serve models, backed up 6:09by the MIT study, ironically. So is Sam 6:12right? Are we in a bubble? I would 6:14actually argue that if he means are 6:16there elements of unfounded hype in AI? 6:18Yes, there are. Absolutely. Is there 6:20froth? Yes. as a wonderful example. 6:23Again, just from this week, I could pick 6:25any number of a dozen examples, but look 6:27at the number of lovable copycats out 6:29there. How many companies do you know 6:31who have put up a little box saying, 6:32"What do you want to build today?" The 6:34latest one is Air Table. I would not I 6:36would not think that Air Table should be 6:38doing that, but they've decided to. With 6:40any gold rush, you get people rushing in 6:42to stake a claim where they think 6:44there's gold. And Lovable has 6:46demonstrated there's gold in vibe 6:47coding, and so now there's a rush there, 6:48right? And there's going to be a lot of 6:49mewoo players. anytime you have value, 6:52you have me too players. That doesn't 6:54mean it's inherently a bubble no matter 6:57what. And I think that people sort of 6:59overindexed on that comment and they 7:01thought there's hype players that means 7:02it's a bubble. Let me tell you, I have 7:03lived through a bubble. That is not the 7:05only element you need for a bubble. I 7:07think that one of the things that we 7:09should balance out with as we look 7:10across like how people got to this 7:12narrative, the things that we've 7:13forgotten, what Sam might have meant by 7:15bubble and what elements are indeed 7:17bubbly in the AI narrative. We need to 7:20also pay attention to what else is going 7:21on. And I think there's there's two 7:23trends that better explain the full 7:25story I've been telling just a bubble. 7:27One is AI is demonstrating real value 7:31and real use cases. And that is 7:32ironically why businesses are leaning in 7:35so hard. The story of the 5% isn't 7:37getting told, but I've seen it. When 7:39organizations get it right, AI is 7:41delivering step change gains. It's 7:43delivering 10x gains across the 7:45business. That is existential. It is 7:47worth betting a lot on. It is why the 7:50organizations that are failing are going 7:51to come back and most of them are going 7:53to try again. They can't afford to miss 7:55this one. The second one is related to 7:57that. We are in a power law game and 8:00power law cost and returns show up 8:02across AO. That means AI is increasing 8:04according to a power law. So it's 8:05increasing exponentially. I talked about 8:07that. It also means you get power law 8:09returns from gambling on AI as a 8:11business. And I gambling is probably the 8:13wrong word. Betting on AI as a business. 8:14Essentially, if you invest in something 8:16and there's a power law return, it's 8:18rational to invest more than you usually 8:20would. And we see that pattern play off 8:21across companies investing in AI, but 8:23also across model makers. Modelm makers 8:26investing a billion dollars in AI talent 8:28or whatever it is, as Zuck did, model 8:30makers investing a huge amount in chips. 8:32All of that is a way of saying we think 8:34there's disproportionate returns on AI 8:37and we are going to keep investing very 8:38very heavily in order to harvest those 8:41returns. Now, I do think one of the 8:43things that's shifted in this game is 8:44that it's harder and harder to catch up. 8:46One of the things that I noticed is that 8:48Apple is trying to figure out how to 8:50recast their narrative in the last week 8:51or two. They need to be seen to be 8:54playing an AI. And so, they had a big 8:56piece that there was a leak. I'm sure it 8:57was a leak that was kind of intentional, 8:59guys. But it is harder and harder to 9:01catch up as we move forward on the AI 9:03frontier. And there are fewer and fewer 9:04labs that are really seriously playing 9:06on the edges of AI. There's OpenAI, 9:08there's Anthropic, there's Google, and 9:10Meta is trying. and XAI is trying. And 9:13other than those, like Amazon has fallen 9:15by the wayside. Microsoft has arguably 9:17just decided to be in the cloud business 9:18and serving AI models business and 9:20that's gone very well, but they're not 9:22really doing something separate from 9:23open AI right now. And part of the 9:26reason for that is that as you get a 9:28power law world with AI, you get 9:31incredible pressure to specialize and 9:33pick your niche because otherwise you're 9:35spending a lot of money for nothing. And 9:36so, ironically, I would argue the fact 9:38that we've seen a winnowing out and a 9:40narrowing of AI model makers in the last 9:43year. It's an argument that people are 9:44actually starting to think about how 9:46they're allocating capital, which is not 9:47something you do in a bubble, and 9:48they're starting to be trying to be 9:50smart about where they play in this AI 9:52world. Microsoft wants to sell the picks 9:54and shovels. They want to sell the cloud 9:55piece. Google wants to sell the cloud 9:57piece. I think AWS does as well, 9:59although less successfully so far. In a 10:01power law world, it pays to invest 10:04heavily if you know your niche. which is 10:06sort of a large strategic insight that 10:08scales all the way out to businesses. 10:09Like you have to know your niche to sort 10:10of be able to invest carefully, 10:12cleverly, and well if you're going to 10:13invest that much. But if you know your 10:16niche, it is rational to allocate 10:18capital heavily. And that's what we see 10:20businesses doing. And so when you lad 10:22this together, some froth, you have 10:24demonstrated real value on use cases, 10:26you have a power law dynamic going on. I 10:28think the way I would put it is that we 10:30are in a world where model makers are 10:32showing exponential gains in model 10:35performance and we are very very early 10:37and seeing how that lands with the 10:38business and that's part of the irony 10:40and the challenge right now in terms of 10:41where this sets us up for the rest of 10:43the year. Listen, I've lived through 10:45multiple bubbles. The one thing you 10:46never see in a true bubble is people 10:48complaining about it being a bubble. If 10:50it really was a bubble, we wouldn't all 10:51be complaining about it. Instead, we 10:53would all be hyping it up. And I think 10:55it's really healthy that we're having 10:56this conversation. It's healthy that 10:57we're asking the question, but when you 10:59look at the narrative overall, I don't 11:01think it I don't think it adds up to 11:03bubble. I think it adds up to a frothy 11:05high capital market where some people 11:07don't know where their their niche is 11:09and they're overallocating in the wrong 11:11spaces. You see some people who are 11:12desperately trying to AI wash their 11:14products and you see real value and the 11:17real value is so disproportionately 11:19helpful to business that people are 11:21doing anything to get it. That's a 11:22complex story. It's going to become more 11:24complex over time. I don't think that we 11:26are going to get into a world again 11:27where we have immediately obvious 11:30chatbot use cases. There are going to be 11:32some immediately obvious AI use cases 11:34for consumers coming. I don't think it 11:36will be in the chatbot. It will be 11:37somewhere else. But we're going to 11:39increasingly see incredibly valuable 11:40business tools come out and I think 11:43we're just at the front end edge of that 11:45piece of the AI revolution. I'm excited. 11:48I'm curious. I am not worried about an 11:50AI winter. Uh, and I hope that this has 11:52helped you recast some of the overall 11:54narrative we're seeing,