That's a question that has been on my mind for a long time. Some voices think so, some voices say that there is still room for plenty of growth.

Let me start of by saying that this is an opinion, based on what I've seen around the market. I won't be talking much about the numbers, because numbers don't mean anything or their story can be skewed in any direction to support a narrative.

I am talking from the perspective of a technical person that also knows a thing or two about businesses.

Yes

My quick 2 cents are that the chances to be at the top of a tech bubble are pretty large, and that should worry us because:

• huge parts of the available capital are funneled into businesses that have no future (mostly related to AI), and they will eventually implode, taking those funds with them.
• this capital being funneled into AI tangent businesses reduces the available capital for other businesses which hinders growth.
• those funds are being spent on crazy useless things, such as .com domains (again, like holy s**t, why would they even need them?). Here I'm talking about how OpenAI spent probably more than 15.5 million USD on chat.com, friend.com being bought for 1.8 million USD (out of their total 2.5 million USD raised funds, so 70% of their funding just went for a domain purchase). And AI.com seems to also be under the OpenAI umbrella for 11 million USD. Read more about crazy domain name spendings here.

Y Combinator, one of the biggest and most influential investment fund seems to be interested nowadays a very large part of their cash in AI related tech. Some of them have very weird use cases and value propositions. Just look at this list and tell me confidently that there are many moneymakers in there

AI bubble will burst. But when/how?

I think that the AI bubble with burst soon. There are so many companies who compete with each other, and they are pouring huge amounts of cash into it. Model training is expensive and even running them as well consumes a lot of resources.

With a 200 USD consumer aimed chatbot, AI claimed it loses money. So for the amount of usage it sees, the AI companies nowadays subsidize a lot of the consumer market with their own funds (or their investors)

This can't be done indefinitely and at some point (I think really soon, sometimes during 2025 or 2026), we will see correction in their business models. I think they will either switch to a pay per use model (or a hybrid that has a subscription component with a certain included usage level + charges for what you use over your plan) while also increasing the per-token pricing that they offer for their APIs.

AI is looking aimlessly for more use cases

There are many potential use cases floated around, but the market didn't provide these companies with reliable and recurring opportunities.

People scramble together all kind of glued together apps and automation that sometimes works, but the "statistical" or "random" nature of the LLMs makes it unsuitable for things that need precision.

And the only use case that it kind of "excels" at is generating content. And the content it generates even got a specific term, "AI Slop".

Many people started to push back against AI generated content as it feels non-human, generic, just mumbo jumbo. And the other produced media types also have a specific "AI fell" that people picked up on.

What is the real value it provides?

I think the only good enough value producing use cases are

• replacing traditional OCR for smaller documents and for operations that don't rely on full precision
• making more malleable image processing pipelines (as the multimodal inputs for most models offers the opportunity to extract data quick from the image itself)
• simple code generation - targeted small and very precise code generation is good enough, as long as you verify it and are able to integrate it back into your project. But for substantial code generation, it hits the mark most of the time.
• ideation and bootstrapping research in various domain you're not really familiar with (it has gotten better due to the fact that many LLM interfaces also offer access to their sources now). For example, creating a calendar and targeting for specific e-commerce products or finding some websites to promote your product on for free.

But these use cases definitely don't justify the hundreds of billions of dollars that are being poured right now in the industry, unless we see a massive breakthrough like AGI, but without being able to "replicate" the human brain complexity as a a bunch of parameters in these models.

So what's next?

We just wait and see. With so many things going around, models seem to specialize in different parameters, and these specializations will make them more suitable for various use cases. For example Mistral AI seems to have specialized their model in the quality of their image generation and speed. OpenAI want to specialize their models in "intelligence", Google's Gemini has an insanely large context size and Claude.ai is still the best model for code generation.

We will probably start to see more specializations and doubling down in certain directions, until we will see some breakthroughs with some of them. For example what if the speed of Mistral AI's model will become the bread and butter, and will make it become the standard model for high volume processing? Or what if Google's Gemini huge context window means that it will become the standard model for large input processing, being able to "read" and process books in one go? Or gigantic PDF files?

But we'll also see some of these AI companies going down, because without widespread adoption and enough money coming in from the market (instead of investors), they won't be able to sustain their current business models indefinitely.