You Ask, I Answer: Is the Generative AI Space Ripe for Consolidation?

In today's episode, Justin asks, Do you think the AI space is ripe for MA, mergers and acquisitions? Oh, yeah. The the space is ripe for mergers and acquisitions or just companies flat out going out of business. And here's why. There are a lot of vendors in the AI space whose value proposition is essentially a wrapper or a user interface or something on someone else's model.

So there's a gazillion different little companies that all have built their company around, for example, OpenAI's GPT-4 model. That model is very capable, it's very powerful, and these and and folks have built a company that puts an interface on top of it that is purpose-built towards one specific set of tasks. And maybe there's some additional value add like document storage. But fundamentally, the underlying baseline model is someone else's model. And so as those models change, if the company has not done a good job of planning for the future, that company gets really far behind really fast.

So maybe you buy some software that's you know blog writing software, um, and it's really just a skin on top of GPT 4 or Claude 2.1 or whoever. If that company did not think through, how do we how do we make our our software abstracted away from the base model, then they have to stay locked into that base model, and when it becomes old, they can't easily adapt to whatever the new thing is. And so they go from being best in class to being last year's news very, very quickly. The AI space is doubling in terms of capacity. Models are doubling in capacity roughly every six months, six to nine months.

So if you were, if you built this bespoke product around GPT-3, for example, that was three years old, you are five or six generations behind. And when it comes to compute power and results delivered, that's a big difference. Your company is essentially as a non-starter compared to what you can do with the foundation models themselves. So a lot of companies have created a lot of value, but in terms of you know what they can get people to pay for, but that may be very transient. Because every release of uh model these days brings new capabilities and makes it easier to replicate things that you might create software around.

For example, suppose you are a company that makes blog writing software, and you your big value proposition is is document storage that you can uh easily use uh your company's documents within this thing. Well, that was fine until October uh November of 2023 when when OpenAI released custom GPTs, and now anyone can take their documents and stuff them in a model and have that information be available and have it be useful and things like that. So I remember I I was watching on Threads when uh the Dev Day uh talk was going on and people commenting, wow, they are just putting companies out of business left and right with every single announcement because every new announcement was building capabilities into the foundation models and the foundation ecosystem that other people have built entire companies around. So, what is the value proposition of that company now that the base system software can do that itself? And there's a lot more coming from the big model makers that are going to imperil a lot of these smaller businesses.

Uh, Andre Karpathi in his recent talk was showcasing how to use language models as a kind of an operating system. Think about that. An operating system for your computer that is based on plain language, even something like Mac OS or Microsoft Windows might be potentially at risk from that. So the AI space is definitely ripe for mergers and acquisitions, is definitely ripe for consolidation. Um, whether that is a company getting acquired or a company just going out of business.

The AI space is right for innovation for every company that's going to go out of business or get devoured, you're probably going to see two or three new companies that are leveraging what is cutting edge right now. For example, there's an open source model called Lava, that is a combination language and vision model that is very, very good and very, very powerful and also very free. You could get a whole generation of people building companies around that model and its capabilities. And because it's open source or open weights, you don't need to pay anyone to use that as long as you are under you know whatever the license terms are for like the Lama 2 derivatives. It's if you have 700 million or fewer monthly users, you can use the model for free.

Language models pro and doing form processing, that's not a terrible stretch, right? Because it still uses language and it uses highly templated language, which should be relatively predictable. Now, doing the math part, that's going to require some app ecosystem around it, something like Langchain or AutoGen or something along those lines. But there's no reason why conceptually that can't exist. If a task uses language, it is it is ripe for a language model to do.

So the space is ripe for MA. The space is ripe for fast transitions. The space is ripe for innovation. And the key message, the key takeaway is you have that opportunity right now. If you got an idea about ways to use generative AI, yeah, probably somebody's working on it, but you can be too.

Because and because the space is so dynamic and so fluid, you can have more than one company that does the same thing. And they, you know, you'll compete for market share, but the opportunities are right now. So get started. Get going. Anyway, really good question.

Thanks for asking. We'll talk to you soon. If you enjoyed this video, please hit the like button. Subscribe to my channel if you haven't already, and if you want to know when new videos are available, hit the bell button to be notified as soon as new content is live.