New Technology / Ai Development
Track AI development, model progress, product releases, infrastructure shifts and strategic technology signals across the artificial intelligence sector.
The Secret Economics Controlling OpenAI, Anthropic, and the AI Boom
Topic
Economic Dynamics of AI Labs
Key insights
- The Cornell equilibrium is an economic construct coined by Antoine Corneau
- In markets with few players, competition occurs on supply rather than price
- AI lab leaders claim they dont think about competition, but they are clearly obsessed with it
- Companies like Microsoft and AWS are in constant competition, adjusting their strategies based on each others actions
- OpenAI and Anthropic are considered deeply unprofitable companies
- AI labs have two businesses hidden within their PNL: training models and inference
Perspectives
Analysis of economic dynamics in AI labs and their competitive strategies.
Proponents of the Cornell Equilibrium
- Highlight the relevance of the Cornell equilibrium in AI market dynamics
- Argue that AI labs are competing on supply rather than price
- Claim that firms are currently unprofitable despite high gross margins
- Emphasize the importance of inference margins in AI business models
- Propose that the oligopolistic structure of AI labs leads to strategic interdependence
Critics of Current AI Market Strategies
- Question the sustainability of AI firms business models given their unprofitability
- Warn about the risks of over-investment in AI resources
- Doubt the long-term viability of relying on a few key players in the market
- Challenge the notion that increased competition will not lead to commoditization
- Critique the assumption that AI labs will evolve into profitable hyperscalers
Neutral / Shared
- Discuss the potential for differentiation among AI models
- Mention the role of venture capital in supporting multiple AI firms
- Acknowledge the ongoing debate about the future of AI model improvement
Metrics
contribution_margin
positive contribution margin
financial health of AI labs
Indicates profitability potential in a competitive market.
you see positive contribution margin
market_structure
oligopoly
competitive landscape of AI labs
Suggests limited competition and potential price control.
a small number of labs, an oligopoly
substitutes_availability
limited substitutes
market alternatives for AI models
Contributes to sustained high pricing for frontier access.
there aren't great substitutes
training_cost
one billion dollars USD
cost to train a model
High training costs can impact overall profitability.
a model got trained that costs a billion dollars last year
revenue_generated
four billion dollars USD
revenue produced by a trained model
Indicates potential for high returns on investment.
this year it produced four billion dollars of revenue
overall_loss
loses money overall
financial status of the company despite individual model profits
Highlights the disparity between model profitability and company financial health.
the company loses money
gross_margin
very positive
current gross margins of the firms
Positive gross margins suggest potential for profitability if costs are managed.
the gross margins right now are very positive
exponential_scale_up_phase
exponential scale up phase
current phase of compute scaling
Indicates rapid growth but also potential for unsustainable practices.
we're still in the exponential scale up phase of compute
Key entities
Timeline highlights
00:00–05:00
The discussion revolves around the Cornell equilibrium, an economic concept that suggests limited market players compete on supply rather than price. It highlights the competitive dynamics among AI labs, particularly the obsession with each other's strategies despite claims of not focusing on competition.
- The Cornell equilibrium is an economic construct coined by Antoine Corneau
- In markets with few players, competition occurs on supply rather than price
- AI lab leaders claim they dont think about competition, but they are clearly obsessed with it
- Companies like Microsoft and AWS are in constant competition, adjusting their strategies based on each others actions
- OpenAI and Anthropic are considered deeply unprofitable companies
- AI labs have two businesses hidden within their PNL: training models and inference
05:00–10:00
The discussion centers on the economic dynamics of AI labs operating within an oligopoly, focusing on their revenue streams and contribution margins. It highlights the high prices for frontier access driven by limited substitutes and the willingness of customers to pay for superior AI models.
- Revenue subscriptions, API usage, and enterprise contracts contribute to positive contribution margin
- Inference margins are considered healthy in the current market
- The market is characterized by an oligopoly where a small number of labs choose supply at the frontier level
- High prices for frontier access are maintained due to limited substitutes and customer willingness to pay
- Developers and knowledge workers are willing to pay hundreds of dollars a month for the best available model
- There is a need for checks and balances due to potential hallucinations in AI models
10:00–15:00
The discussion focuses on the competitive dynamics among AI labs, particularly their strategies for acquiring the best resources and the implications of potential market shifts. It also addresses the uncertainty surrounding the future of model improvement and the possibility of transitioning to a more competitive market structure.
- They want the best tokens, deep research reports, and code, and are willing to pay for it
- Open AI and inthropic are keying off each other in the market
- Having an API business gives leverage as models improve
- Inthropics critique was their constant need to train better models
- Dario is discussing the potential end of exponential model improvement
- There is uncertainty about creating super intelligence
15:00–20:00
The discussion focuses on the economic dynamics of AI firms, highlighting their limited investment in R&D and the high profit margins on marginal costs due to efficient inference. Despite positive gross margins, the three leading firms are currently unprofitable as they scale up their compute resources exponentially.
- The industry has a small number of firms that can invest limited amounts in R&D
- The profit margins on marginal costs are high due to efficient inference
- There is competition among firms, but they behave independently and rationally
- Currently, three leading firms are not making profit despite positive gross margins
- Companies are in an exponential scale-up phase of compute
- Training a model can cost a billion dollars, but it can produce four billion dollars in revenue
20:00–25:00
The discussion revolves around the competitive dynamics and strategic interactions among AI labs, emphasizing the concept of Nash equilibrium in their operations. It suggests that despite increased competition, these labs are likely to remain profitable and avoid commoditization.
- We block each other
- Nobody likes to be second choice
- The best result comes from everyone in the group doing whats best for himself
- Nash equilibrium
- Lots of game theory going on in the AI wars right now
- Theres a fair amount of risk