The AI Bubble: How to pick the winners before it bursts

It is a continuously repeating phenomenon in history that investors grossly overvalue assets in a ‘hyped’ market because of speculative forces. Whenever this happens people tend to lose a lot of money.

Considering fundamental economics, the current value of AI stocks is not justified. More specifically, the stocks of AI chatbot companies are valued as if their future profits will be very large, but the current market for AI chatbots is very competitive, meaning no one party will be able to extract high profits.

Human nature and FOMO drive stock prices even more than rational analysis; a wise investor should learn from history and adjust their decision making accordingly.

The problem with the AI bubble

In the late 1990s to early 2000s investors poured excessive amounts of money into tech startups due to hype, FOMO and low interest rates. The internet, a new technology full of potential and unproven business models, had everyone scrambling to invest in an internet company or start one themselves so as to get in on the action. Does this sound strangely familiar in 2025?

What happened next was investors realised that the internet companies’ valuations were far above their earnings potential, and the value of the Nasdaq index crashed 77% from its peak of 5,048 units on March 10, 2000 to 1,139.90 units on October 4 2002.

The reason this could happen in the first place is because investor excitement about tech stocks became disconnected from their realistic earnings potential - smart investors should learn from the mistakes of the past to avoid losing money in the same way.

A disambiguation on the term “AI”

The media do a very bad job of explaining AI to non-technical audiences. This word has become overused and now has a very nebulous meaning. In this article, we talk about the focus of the current AI hype, which is so-called Large Language Models (LLMs), an umbrella term for the algorithms which are powering AI chatbots such as OpenAI’s chatGPT, Meta’s Llama and Anthropic’s Claude.

LLM is the term used to describe the application (i.e. we are using an AI algorithm that processes language, and the size of the AI model is very large) rather than describe the structure of the algorithm, where you might see a word like transformer model instead (as seen in the acronym chatGPT, GPT being Generative Pre-Trained Transformer, but this is AI jargon).

To complicate things even more, many companies like OpenAI whose main product was originally LLMs, have also developed so-called multi-modal models. Multi-modal models refers to integrating sound, image and sometimes even video processing into the original LLM, for example if one can input text into an AI model telling it to generate a Studio Ghibli-style image of a cat, this would be a multi-modal model.

The reason we are making a distinction between LLMs and other types of AI is that LLMs are the most visible application at the moment, and simultaneously the most inflated bubble. Different types of AI algorithms are being used in many different fields like robotics, drug discovery, and autonomous driving, but these are not receiving the same amount of media attention as LLMs are.

How does an LLM Chatbot make money?