(This post was written originally by a human, me. This English version was translated with AI from the original.)
Before we get to our analysis, let’s briefly recap the events.
The first event that triggered these debates was the release of ChatGPT-5 and the disappointment it caused for many people. This created in some the impression that technological progress in this field had reached its end and that we had entered a period of stagnation. This period, called the AI Winter, has happened many times before in the nearly 70-year history of AI research. In those periods, AI research hit dead ends, hopes and interest waned, and most importantly, funding dried up.
The second event that fueled the debates was the sudden one-day drop in technology and AI stocks. The drop lasted only a day, but it was enough to revive bad memories of the dot-com bubble that burst in the early 2000s. Major newspapers simultaneously started questioning an AI bubble.
The third event accompanying all this was that technology leaders in AI, like Sam Altman, suddenly changed their tune—in other words, they began speaking about AI in a more cautious, expectation-lowering tone. This was particularly striking, because it’s unusual for a CEO—whose job includes promoting the company and raising expectations—to try to lower expectations.
Before we pick a side in this debate, we need to define what “bubble” is. As we’ll see, participants in the debate are talking past each other using different definitions.
Rather than going straight to a dictionary, I’ll try to offer my own definitions:
Definition 1 (Sociological): Fashion. A passing fad. Something that draws enormous public attention for a period but soon goes out of fashion and is forgotten.
Definition 2 (Financial): In economics, a good’s price becoming excessively overvalued in an unrealistic way. Demand surges excessively. Eventually, when reality asserts itself or demand falls, the price drops suddenly back to normal.
Definition 3 (Technological): Overstating the value and benefits of a new product, technology, or scientific discovery.
Definition 1 is about perception. Masses talk a lot about certain things in certain periods and soon forget them. In light of Definition 1, AI is definitely a bubble. Because yes, people, the media, and social media are currently paying a lot of attention to AI. After a while that attention will definitely fade.
The second and most commonly known definition of a bubble concerns economics—more precisely, finance. Whenever bubbles come up, economists recite by heart, as if they had seen it with their own eyes, the historical episode called the Tulip Mania. But when it comes to history, it’s worth treating economists with some caution, because they tend to use and bend history to support their own theories.
The truth is we don’t need to go back to 17th-century Holland to believe in the existence of bubbles. These are things that happen every day within the natural workings of markets. Markets try to price things whose value we don’t really know, and they do this in ways open to all kinds of speculation. The fact that these rises and falls are sometimes sudden and large, and cascade into crises, is simply a consequence of the nature of the game.
In light of this second definition, is AI a bubble? Unfortunately, we have no way of knowing that today. It’s a fact that astronomical amounts are being invested in AI. For example, according to The Guardian, in the first seven months of 2025, large American tech companies invested $155 billion in AI. How much and when these investments will pay back—I honestly don’t think anyone knows.
The dot-com bubble is seen as a textbook example of this, yet in the long run it’s debatable whether it was a bubble or an opportunity. Take pets.com, one of the era’s most well-known examples. The company was founded at the end of 1998 to sell pet products. Over its two-year life it took in a total of $50 million in investment and sold most of its shares to Amazon. It spent a significant portion of that investment on marketing, including a $1.2 million Super Bowl ad.
For years, pets.com was a punchline. The main reason was the idea that selling pet food online was such a bad business idea it bordered on stupidity. There was no sense in buying a heavy product like a bag of kibble online to ship by mail. Investors, it was said, had fallen for a bubble and poured money into a bad idea. Yet by 2024, in the U.S. alone the online market for pet food and products had reached close to $30 billion annually. Amazon’s own value fell from $28 billion when the bubble burst in 2000 to $5.5 billion by year’s end, and in the following years receded to just a few billion. Today it’s worth $2.4 trillion.
Such massive capital flows, of course, don’t directly concern us ordinary mortals. What concerns us is hidden in the bubble’s third definition. In fact, this is the “bubble” most people in these debates are talking about. Let’s recall:
Definition 3 (Technological): Overstating the value and benefits of a new product, technology, or scientific discovery.
Some familiar examples spring to mind here. The metaverse is the first. Blockchain, NFTs, cryptocurrencies are others. These are basically “buzzwords,” fashionable terms used—especially in the startup world—to attract investor interest. Their common point is that the media inflates them for a while as the “technology of the future,” then they’re suddenly forgotten. Yet the technological innovations they rest on are very real. The only problem is that their value and use can’t be predicted well enough.
There’s also something else about these bad examples. All of them were used in the form of Ponzi schemes to raise money from ordinary people. They all promised easy riches. Under the name “metaverse,” people were sold virtual land. Under the name “NFT,” virtual “artworks” found buyers at high prices. People invested in “projects” under the banner of blockchain that they didn’t understand—or thought they were investing in. As with any Ponzi scheme, some made a lot of money, but the majority lost theirs.
AI, of course, has nothing to do with these money-trap examples. But technological bubbles aren’t limited to these.
It’s time to recall Amara’s Law in this piece.
“We tend to overestimate the short-term effects of a technology and underestimate its long-term effects.”
—Roy Amara
We call it a law, but of course it’s just an observation—one that matches reality for all kinds of technological innovation. For every new product or technology, people’s, the media’s, and money’s attention gets caught up in excessive short-term expectations. In the long run, the opposite happens: interest wanes while the technology’s concrete effects can overshadow even the initial expectations. In short, public attention and a technology’s real utility generally proceed independently of one another.
This is so well known that the progress of new technologies is regularly tracked and published by the consulting firm Gartner under the name “Gartner Hype Cycle.” Naturally, all tech investors and CEOs are well aware of these time-dependent swings in expectations.
Finally we’re getting close to where we can find the real answer to our question. If we think of AI independent of all the media attention, financial swings, and expectations, is it a passing fad? Or truly a revolution?
The second question: can this revolution continue? Or will progress be interrupted and the field of AI go back into hibernation?
Answering these questions may look like fortune-telling, but they’re actually questions with a technical side, and we have enough clues.
Yes, today’s AI technologies are a real revolution. A series of inventions created inflection points that first spread across all engineering fields and then into every area of life. In my view the main inventions, in order, are: 2012 Deep Learning, 2017 Transformers, and 2020 GPT-3. Beyond those, a new ecosystem—including hardware, software, and compute—has emerged. We need to think of this as a new computing paradigm.
Will inventions and progress continue? It’s impossible to know. But even in its current state, we have technology in hand that can shape the world over the next 10 years. In other words, even if all AI research and investment in labs stopped today, the mere productization process would be enough to keep us busy for years.
When people hear AI, most think of ChatGPT or similar applications. They see these apps from a consumer’s perspective, as finished products. They value the entire field of AI based on whether this product is useful to them or not. But ChatGPT (Claude, etc.) is just one application of a foundational technology under development. It started as a kind of technology demo, but quickly became one of the world’s most popular apps. For many reasons—collecting real user data, learning user behavior, and early entry into the global market—it’s offered free to the whole world.
This foundational technology is being developed to form the backbone of many more comprehensive systems. It is entering everything that uses technology—from banking to defense, from manufacturing to marketing. The first quarter of the 21st century was defined by “Software is Eating the World.” The fact that the world’s largest companies are information technology companies and that software firms have taken over every sector, relevant or not, was because of this. In the second quarter of this century, software will continue to eat the world by encompassing the technology we today call AI.
This same foundational technology already seems to have shaken software development to its core. A computer’s ability to write meaningful code on its own—i.e., in a sense to program itself—is an incredible thing that shouldn’t be underestimated. For us humans, code written in programming languages was the only way to make computers do what we wanted. The goal was not to write code, but to communicate with computers. That has now changed. It may look like we’re in the crawling stage, but a complete paradigm shift in software development tools and processes seems inevitable. For now, the only thing we can do is have AI write code in the (soon-to-be-legacy) programming languages designed for humans. Even that is spreading at incredible speed.
LLMs (Large Language Models) are named language models, but in reality language here is just a tool—a tool we use for everything from communication to reasoning, from transmitting knowledge and ideas to making decisions. That tool is no longer our exclusive domain. There is now an entity in the world that uses language better than we do. Moreover, an LLM’s language ability is only its outward face—the way it communicates with the outside world. No one has yet fully identified what other capabilities it contains within.
When we say AI, people think of text-based systems, but the essence of the technology that’s been invented imposes no such limitation. Text, images, sound, motion, music, mathematics, heartbeat, DNA, molecules, telescopes… anything that can be converted into data (i.e., everything) can, in theory, be the system’s input or output.
LLMs—the machine-learning models we today call AI—don’t have to do everything by themselves. For example, their doing simple arithmetic wrong or miscounting the letters in the word “blueberry” are not as important as people think. Computers are already perfect machines for performing arithmetic or counting letters. These are not the things AI is expected to do. Engineers actually know this well, but since it’s the first mistake ordinary users fall into, I felt the need to add it. Thankfully, popular products like ChatGPT are rapidly gaining features for using tools and working together with traditional systems like agents, function calling and the like.
We could extend the list. The point I want to make is that AI is real and it’s here to stay. Tomorrow we’ll slap different names on what we today call AI. But it will keep entering our lives more each day—so much so that we’ll even forget it’s there. A quiet, underground revolution will take place. There’s no rule that this revolution will make humanity better. We might be affected negatively. But because we can adapt to everything quickly, we won’t notice that either.
An arrow has already left the bow. No one knows how far it will go or where it will land. There’s no point in arguing about it or betting on whether the arrow is a bubble.