Nate's Newsletter · Tech & AI
TIER 4 Thu, 21 Aug 2025 13:03:06 +0000
Watch now (12 mins) | There's been lots of talk of an AI bubble this week--here's why that's wrong, what people are missing, and what to expect next ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ | | ---|---|--- | | | Forwarded this email? Subscribe here for more --- Thanks for supporting Nate's Newsletter! This post is free and clear for all subscribers, just like the sky (but not in Seattle--it's always cloudy here). * * * --- --- | | Watch now --- # AI Bubble? Why the Doom Narrative is Wrong ### There's been lots of talk of an AI bubble this week--here's why that's wrong, what people are missing, and what to expect next | | Nate --- | Aug 21 --- | --- --- | | | --- | | --- | | --- | | --- | | READ IN APP --- This week I've been watching the AI narrative collapse in real-time, and it's fascinating for all the wrong reasons. Sam Altman made an off-the-cuff remark about bubbles. Meta is laying off some in their AI division. AI tech stocks are having a down week. GPT-5 disappointed lots and lots of people. The prophets of doom are having their moment. But as fun as it is to ride the narrative rollercoaster, the prophets of doom are incorrect. Things aren't as gloomy as people make out, and I think there are solid reasons why it looks doom-y at first glance anyway. This note lays out why I think we're hearing this narrative everywhere, what people are forgetting about how AI works, and where we go next. Let me show you what everyone's missing. Subscribers get all these newsletters! Upgrade to paid * * * # The AI Bubble That Wasn't: A Story of Misread Signals The headlines are everywhere this week. Sam Altman admits we're in an AI bubble. Meta freezes AI hiring after its lavish spending spree. MIT finds 95% of enterprise AI projects are failing. The prophets of doom are having their moment. Last night on X, I watched the narrative crystallize in real-time. Tweet after tweet about AI winter or about AI hype being over or about impending stock collapse. The sentiment has shifted so fast it's giving people whiplash. But here's what's actually happening, beneath the noise. ## The Perfect Storm Four things are converging right now to create this bubble narrative, and if you don't see how they fit together, you'll miss what's really going on. First, people desperately need a new story. We spent months building up GPT-5 as the second coming. When it launched last week and users found it cold, impersonal, even inferior to GPT-4 in some ways, the backlash was swift and brutal. Sam Altman had to issue that rare public apology--"We totally screwed up"--and now thousands of users have signed petitions demanding their old models back. The hype pendulum had swung too far; now it's swinging back with vengeance. Second, Meta's AI division restructuring hit the news cycle at exactly the wrong moment. After months of throwing astronomical compensation packages at anyone who could spell "transformer," they suddenly freeze hiring and announce a reorganization into four separate units. The optics are terrible, even if the reality is more complex. Third, Sam himself uttered the magic word at that dinner with reporters on Monday. "Are we in a phase where investors, in general, are overexcited about AI? My opinion is yes." He went on to compare it to the dot-com era, warning that "someone's going to get burnt" on overvalued AI startups. When the CEO of OpenAI says bubble, the world listens. Fourth, MIT's study lands like a bomb in the middle of all this. 95% of corporate generative AI pilots failing to deliver meaningful returns. Companies have poured $35-40 billion into generative AI with almost nothing to show for it. The timing couldn't be worse for the AI optimists. The market's responding predictably. Tech stocks are down. Nvidia dropped 3.5% this week. The narrative has taken hold. ## What We're Missing But narratives seduce us precisely because they simplify. They transform messy reality into clean stories. And this bubble narrative is missing five crucial facts that completely change the picture. **First:** The chatbot use case is saturating, and Sam's actually being honest about it. In the same interview everyone's quoting about bubbles, he said something more interesting: ChatGPT has 700 million weekly users and "pretty soon billions of people a day will be talking to ChatGPT." The chatbot interface is good enough. Making the model smarter doesn't fundamentally change that experience anymore. This isn't failure; it's market maturity in one specific application. **Second:** Progress hasn't stopped; it's moving to places most people can't see. Take the debate raging on X right now about whether GPT-5 Pro just created a genuinely new mathematical proof. The consensus emerging is yes, it's real innovation, but it's different from human innovation--brute force calculation through a problem space rather than intuitive leaps. The nuance of that distinction is lost on most observers, but it matters enormously for understanding where AI is headed. **Third:** On any benchmark that isn't saturated, progress continues exponentially. METR's autonomy evaluation is my current favorite because it hasn't topped out. It simply asks: can an AI complete a task that takes a skilled human X hours, successfully at least 50% of the time? We keep doubling capability every few months. No plateau. No slowdown. **Fourth:** We're still desperately under-allocated on chips, and this is the tell that everyone's missing. In that same dinner where Sam talked about bubbles, he admitted: "We have better models and we can't release them because we don't have the capacity." Think about that. OpenAI has built smarter models that they literally cannot deploy because demand for their current models is too high. Anthropic faces the same constraint. If AI interest was cooling, would we have a chip shortage? **Fifth:** Meta's restructuring isn't retreat; it's preparation for the next phase. Their new chief AI officer's internal memo, which leaked yesterday, doesn't sound like capitulation: "Superintelligence is coming, and in order to take it seriously, we need to organize around the key areas that will be critical to reach it." That's not the language of a company backing away from AI. ## Reading the MIT Study Right That MIT study deserves closer examination because everyone's interpreting it backwards. Yes, 95% of enterprise AI projects are failing. But think about what that means: 95% of companies are trying so hard to implement AI that they're running pilots. The demand signal is deafening. More importantly, look at what the 5% who succeed are achieving. The study mentions startups jumping from zero to $20 million in revenue within a year. Established companies seeing what they call "exponential gains" in productivity. The failures aren't about AI's limitations--they're about implementation. Companies trying to build their own tools instead of partnering. Focusing on the wrong use cases. Lacking proper data infrastructure. Missing the cultural change management entirely. If AI didn't work, companies would try it once and quit. Instead, they're failing and immediately trying again, because they've seen what happens to the ones who get it right. That's not bubble behavior; that's recognition of existential stakes. ## The Power Law Game What the bubble narrative completely misses is that we're playing a power law game now, and power law dynamics explain everything weird we're seeing. In a power law world, the companies that get AI right don't see 20% improvements. They see 10x transformations. When those are the stakes, it's rational to keep investing even if your success probability is low. The expected value is still positive. This extends to the model makers. When intelligence improvements follow a power law, and when capturing even a small slice of the AI market could mean trillion-dollar valuations, spending billions on talent and compute isn't irrational exuberance--it's rational capital allocation in a winner-take-most market. The consolidation we're witnessing--fewer labs seriously competing at the frontier--actually signals market maturity, not bubble dynamics. Amazon has quietly ceded the model race (at least it looks like it). Microsoft has chosen to be the infrastructure provider rather than compete with OpenAI directly. Apple is scrambling just to stay relevant. This isn't the behavior you see in bubbles, where everyone piles in regardless of competitive advantage. This is strategic positioning by players who understand the real costs. ## The Paradox of Bubble Talk I've lived through real bubbles. Dot-com in 2000. Housing in 2008. Crypto in 2021. Here's what I know: the one thing you never see in a bubble is people talking about the bubble. In a real bubble, skeptics are dismissed as dinosaurs. Your taxi driver is giving you stock tips. Your grandmother wants to know how to buy NFTs. Everyone's a true believer, and the few warning voices are drowned out by the chorus of "this time is different." Right now? We're having widespread, serious discussions about whether AI is overhyped. The Financial Times is running skeptical pieces. Even the bulls are adding caveats. That's not bubble psychology. That's the messy middle of a real transformation. ## What's Actually Happening We're not in a bubble. We're in a transition from the easy, visible wins to the harder, more valuable applications. The chatbot era gave everyone something tangible. My mom uses ChatGPT. That amazement phase is ending, not because AI stopped improving, but because we've normalized it. Now we're moving into enterprise automation, agentic systems, specialized applications. These are less visible but more transformative. A consumer can't see when a pharmaceutical company uses AI to design new drug compounds 10x faster. They can't see when a logistics company optimizes routing to save millions in fuel costs. They can't see when a law firm automates document review to handle cases that were economically impossible before. But these invisible transformations are where the real value lives. Yes, there's froth. Every startup is slapping a "What do you want to build today?" box on their homepage, desperately copying Lovable's success. Airtable just did it, and frankly, they should know better. But froth isn't a bubble. Froth is what happens when genuine value creation attracts imitators. ## The Real Story What we're witnessing isn't a bubble bursting. It's an industry growing up. The shift from "AI will change everything tomorrow" to "AI will change everything, but it's going to be messy and take work." When Sam says OpenAI plans to spend trillions on data centers, he's not talking bubble talk. He's talking about meeting genuine demand they currently can't serve. The companies pouring billions into AI infrastructure aren't drunk on hype. They've run the numbers, and even with high failure rates, the expected value is positive. The narrative wants to be simple: AI was overhyped, now reality is setting in. But the reality is more nuanced: AI is delivering transformative value, but only for those who figure out implementation. The technology works. The challenge is organizational, cultural, strategic. We're not in a bubble. We're in a gold rush where most prospectors are failing because they don't know how to pan for gold, not because there's no gold. And unlike most gold rushes, this one is just beginning. The real AI transformation--the boring, valuable, enterprise stuff that actually changes how the economy works--is only now getting started. Be skeptical of the hype. But be equally skeptical of the doom. The truth, as always, is messier and more interesting than either narrative suggests. I make this Substack thanks to readers like you! Learn about all my Substack tiers here Upgrade to paid | | ---|---|--- _You're currently a free subscriber to Nate's Newsletter. For complete strategic insights that'll accelerate your career, treat yourself to an upgrade._ Upgrade to paid --- | | | Like --- | | Comment --- | | Restack --- (C) 2025 Nate 548 Market Street PMB 72296, San Francisco, CA 94104 Unsubscribe