Industrial organization insights on antitrust, digital platforms, and competition in ecosystem markets.
blog.econworks.com (https://blog.econworks.com?utm_medium=podcast)
How Cartels Really Work, Episode 5: The Race to Confess
A leniency program appears to generate a straightforward game theory problem for cartels. Suppose some firms agree to collude. Each wants to make cartel profits, but each knows that another cartel member may go to the government first and seek leniency. So what are you waiting for?
If everyone wants to confess before everyone else, then backward induction makes the confession happen earlier and earlier. The logical prediction is surprising: the cartel should not form, in the end. But cartels do exist.
In Episode 5 of How Cartels Really Work, we examine the gap between that simple prediction and reality. We investigate why deterring is so difficult to observe, why cheating and leniency both erode trust in cartels, and what happens when a profit-maximizer AI is free to choose between competing, colluding, cheating, and seeking leniency.
The final question may be the most intriguing: Could an AI programmed to maximize profits be the perfect cartel partner—or the perfect cartel traitor?
Read the full article and graphic analysis: https://blog.econworks.com/p/cartels-episode-5-the-race-to-confess?r=562wri (https://blog.econworks.com/p/cartels-episode-5-the-race-to-confess?r=562wri)
Explore more visual economics content: https://econworks.com
YouTube: https://www.youtube.com/@EconWorks-d3e
Substack: https://blog.econworks.com
This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe (https://blog.econworks.com/subscribe?utm_medium=podcast&utm_campaign=CTA_2)
2 Oct 2026
Why NVIDIA Is Betting $13 Billion on Open AI Models
NVIDIA sells the computing power behind much of today’s AI boom. OpenAI and Anthropic are major sources of that demand. So why would NVIDIA spend nearly $13 billion acquiring Hugging Face, a platform that helps open models compete with them?
This episode uses a simple three-layer AI structure—chips, model development, and developer toolkits—to examine the economics of the deal.
Closed-model firms such as OpenAI and Anthropic integrate the model and toolkit layers. NVIDIA and Hugging Face would instead connect the chip layer below to the toolkit layer above, while leaving model development open to many competitors.
The valuation provides another clue. NVIDIA is paying roughly 86 times Hugging Face’s reported annualized revenue, far above ordinary software multiples and well above the roughly 15-times multiple implied by the Cursor transaction.
Perhaps NVIDIA is buying Hugging Face for reasons other than the revenue Hugging Face earns. It may be buying the additional chip demand—and reduced customer dependence—that a stronger open-model ecosystem can create.
Read the full article and graphic analysis: https://blog.econworks.com/p/why-nvidia-wants-openai-to-have-competitors?r=562wri (https://blog.econworks.com/p/why-nvidia-wants-openai-to-have-competitors?r=562wri)
Explore more visual economics content: https://econworks.com
YouTube: https://www.youtube.com/@EconWorks-d3e
Substack: https://blog.econworks.com
This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe (https://blog.econworks.com/subscribe?utm_medium=podcast&utm_campaign=CTA_2)
30 Sept 2026
Why NVIDIA Wants OpenAI to Have Competitors
NVIDIA sells the chips that companies like Anthropic and Open AI use. Now it is spending almost $13 billion to acquire Hugging Face, one of the most important platforms for open AI models. Is NVDIA trying to prop up the competition against OpenAI and Anthropic? The economics are more compelling than just a foreclosure story. In this episode, we examine two forms of vertical integration through the lens of a simplified AI supply chain, consisting of **chips → model development → developer toolkit**. OpenAI and Anthropic package models with their developer tools. Instead, NVIDIA and Hugging Face would combine the chip layer and a common toolkit but keep the model layer open for competition. One telling clue we also look at: NVIDIA is paying about 86 times Hugging Face’s reported annualized revenue vs. about 15 times the annualized revenue for Cursor. That implies that NVIDIA might be buying something that Hugging Face doesn’t directly monetize: a stronger open-model ecosystem that could drive more demand for NVIDIA compute. This leads to a counterintuitive possibility. NVIDIA may want OpenAI and Anthropic to succeed—but it may not want to depend on them.
Read the full article and graphic analysis: https://blog.econworks.com/p/why-nvidia-wants-openai-to-have-competitors?r=562wri (https://blog.econworks.com/p/why-nvidia-wants-openai-to-have-competitors?r=562wri)
Explore more visual economics content: https://econworks.com
YouTube: https://www.youtube.com/@EconWorks-d3e
Substack: https://blog.econworks.com
This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe (https://blog.econworks.com/subscribe?utm_medium=podcast&utm_campaign=CTA_2)
Host of EconWorks Podcast?
Claim your podcast to manage its listing and keep your show details accurate.
Pod Engine is an independent podcast discovery and analytics service and is not affiliated with or endorsed by this podcast. Artwork and show content belong to their owners. Full legal notice.
