People often ask me what I think about generative AI. It’s awesome, and terrible, depending on your point of view. From a technical perspective it’s a huge breakthrough that will have lasting impacts. The things we can now do with tools like ChatGPT and Stable Diffusion are pretty remarkable. But technology revolutions often impact shake up the economics of various industries and business models, and by that measure, generative media is a doozy. As someone who has been writing about AI, building AI startups, and investing in AI for the past 8 years, I’ve had a lot of investors reach out to discuss these changes, and I thin the vast majority of them are thinking about the economics of generative media the wrong way.
I think generative AI, the way most investors are currently playing the market, is a bad bet. Let me explain why, and where I think the real value lies.
My perspective on investing is that long term value creation is driven by growth, innovation, and defensibility. That last point is the big one - defensibility. As technology makes doing more stuff faster and easier, it’s increasingly difficult to find areas of long-term defensibility in business models. In particular I don’t see much in generative AI. The platform side - the creation and maintenance of these models has some defensibility in the near term, the next few years, but most people can’t play there. That market will be an oligopoly. It takes tens of millions of dollars to collect data sets and train these models and right now that is only available to the large tech companies. In the long run, new chips may change that equation but, that’s still a few years away.
The key position investors seem to be taking is that “context layers” that take these generative tools and put them into some point solution of a workflow is the place to make a bet. They believe the big tech companies will own the foundation models and so, contextual layers on top of those models that enable workflows like copywriting or image editing or whatever is the way to go. Dozens of generative media point solutions are getting funded.
My opinion is that these business models are not defensible for two reasons. First, there will be too many players because the barriers to entry are low and that drives a competitive dynamic that is unfavorable to investors. Second, they risk competition with the foundation models themselves as those models improve.
So where do you play in this market? My framework for thinking through this is simple. When the cost of something trends towards zero because of new technology:
You will get an explosion of that good.
That good will decline in value and defensibility
The economic complements to that good that see increased demand as a result of the explosion in the original good, will be the place to invest.
To give an example, think about printed news content before the web. You had to have a printing press and a distribution network for your paper. These were expensive to build and maintain and thus local papers had little competition. Then came the web. Now anyone could start an online publication! And for the next two decades, there were very few success stories because that market was too competitive. Margins for news organizations dropped, and were difficult to recover.
When there was suddenly an explosion in content from the web, where was the bottleneck? Search. Finding good stuff in a sea of content became the new economic choke point, and Google became worth more than all of the other newspaper companies combined. The new media organizations that started because it was easy and cheap to start one, that maybe were vertically targeted or, the equivalent of our “context layer” to generative media, made a little profit here and there but, by and large it went elsewhere. I think generative media startups will see a similar trend.
If that is true, where does the value accrue as a result of the generative media explosion? When it becomes really easy to create longer form content and imagery, I believe there are two things that go up in value. The first in orchestration. Coordination layers that help humans deal with the deluge of creative assets they will soon have to navigate will go up in value because doing it well will be difficult.
The second is trust and protection of things like Brand. When machines allow your team, or anyone on the internet, to easily create tens of thousands of images related to your brand, how do you monitor and protect that? How do you insure brand consistency in your own workflows? How do you spot imposters and fakes when they are easier than every to create?
These will be difficult problems to solve, but they will be profitable ones. I’m convinced enough of this that I’ve launched Nova, which has been over a year in development working on exactly these issues. If you run a marketing team and are interested in using generative media, we’d love to chat.
Thanks for reading, and as always I’d love any feedback you have if you disagree with any of these points.
It's more like $2.5M for data gathering, pre-processing, acquisition of ML Ops tooling, and training of a ~100B parameter model with ~1,000B tokens.
Generally most these companies will fail, but some will succeed.
Rob, super interesting insights. What’s an example of "Coordination layers"? Is that like Digital Asset Management type of tools?