When ChatGPT caught the public’s attention in late 2022, a familiar question followed: is this a bubble? With valuations in the tens of billions and companies reporting heavy annual losses, the skepticism was reasonable. Standing in 2026 and still asking the same question, however, risks missing the change that has already happened.
The comfort of calling it a bubble
Declaring that “AI is a bubble” is reassuring. If it is a bubble, your skills are safe, your career plan still holds, and you do not need to become a student again at 40 or 50. The idea offers the comfort that the world has not changed, only grown noisier.
History suggests otherwise. The dot-com bubble burst and thousands of companies disappeared, but the internet did not. It went on to reshape every industry. People who stopped taking the internet seriously after 2001 found themselves outside the largest economic shift of the following decade. The bubble ended; the transformation continued.
Prices are falling, not rising
Something different is happening with AI. In a classic bubble, prices rise while the underlying quality does not. With AI the pattern is reversed: models are getting cheaper, more capable and faster. According to Stanford’s 2025 AI Index report, the cost of querying a model that performs at the level of GPT-3.5 fell more than 280-fold between November 2022 and October 2024. Tasks that looked impossible three years ago now take minutes.
When a technology becomes cheaper and more powerful at the same time, usage does not shrink. It multiplies. Economists call this the Jevons paradox, after the 19th century observation that more efficient steam engines led to more coal consumption, not less.
The companies spending the most, including OpenAI, Google, Microsoft and Nvidia, are not betting on pure speculation. They are organizations that make decisions on data, trends and long-term strategy. Several of them also report using current models to help build the next generation, an early version of a self-reinforcing development cycle.
An asymmetric bet
The real question is not whether AI is a bubble. It is how the risks balance. If you spend time learning to use AI and it turns out to be a bubble, you have lost a few hours and a few dollars. If it is not a bubble and you waited, you have lost months or years of advantage. The downside of acting is small; the downside of waiting is large.
The most dangerous state is the one where everything feels fine. The salary arrives, the work gets done, life proceeds as normal. That comfort is the trap. Discomfort drives change; ease produces inertia. Many people are not moving today simply because nothing hurts yet.
Change the question
Instead of asking whether AI is a bubble, ask what you can do with it. Whether you are a lawyer, work in finance or marketing, or manage a team, try running a real piece of your own work through an AI tool. The first attempt will not be perfect, but it will be a start.
AI on its own does not deliver the benefit. The combination of a skilled person and AI does. Those who start early compound their advantage month by month. Three years from now, you will either be glad you started on time or wish you had.
Major shifts tend to pass through three stages: first a joke, then denial, then an unavoidable reality. The evidence suggests we are somewhere in the second stage. The practical choice is whether to be ready for the third.
