AI Labs and Movie Studios
Recently, there’s been a lot of talk about what the future holds for the AI labs:
- Is intelligence a commodity?
- Will value accrue to the labs or the app layer?
- Are the foundation labs just the utility companies of tomorrow?
The AI labs are not utility companies, they are movie studios.
Underwriting a $100M bet
Both OpenAI and Marvel:
- Invest $100s of millions upfront
- Spend about 1 year filming the movie (or training the model)
- Create something that makes most of its revenue in the next 3-24 months
- Are high stakes about opening weekend (because it drives demand)
The last Avengers movie cost $200-400 million. OpenAI spent $63-79 million training GPT-4. And Google dropped $191 million on Gemini Ultra.
Once the movie is out, they earn a lot of their revenue early. First at the box office, then through on-demand, and finally as a long tail on streaming.
The new new
While both movie studios and AI labs depend on others to stream their content, because (1) there are only a few AAA movies and models, and (2) because the movies and models are not exactly the same (regardless of how similar the plots of the last two Marvel movies are), people don’t see them as commodity and they can continue to command a premium.
Individual movies do get cheaper over time—what costs $15 at the theater eventually becomes a $3 rental. But people don’t stop going to theaters. They want to see the new releases, not last year’s movies.
Because of that, AMC can’t offer you the old Avengers in place of the new one. And AI apps can’t offer me the old GPT in place of the new one either. They both need to pay the studio, and that fundamentally limits their margin.
When streaming services became studios
At some point, the margin becomes a problem and your biggest customers become your competitors. Netflix and Apple both started studios when they realized they couldn’t depend on someone else’s content.
The response of many studios has been to become streaming services themselves; Disney Plus is even one of the more popular streaming services.
AI is following the exact same playbook, the person creating the content seems to be keeping the lion’s share of the value, and the app layer doesn’t like it, so—if they can afford the CapEx—they’re making their own models.