StarZero All shows Try the podcast clipper →

Feb 18, 2026 · 22m · tbpn

The Secret Economics Controlling OpenAI, Anthropic, and the AI Boom

Packy McCormick · 13m spoken Jordi Hays · 2m spoken Dario Amodei · 2m spoken John Nash (Character in 'A Beautiful Mind') · 37s spoken John Coogan · 25s spoken Martin Hansen · 14s spoken Neilson (Character in 'A Beautiful Mind') · 14s spoken Sol (Character in 'A Beautiful Mind') · 4s spoken Bender (Character in 'A Beautiful Mind') · 2s spoken Dwarkesh Patel · 0s spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

Packy McCormick and Jordi Hays analyze the competitive and financial landscape of the artificial intelligence industry through the lens of Cournot equilibrium and game theory. They demystify frontier AI economics, exploring how training depreciation, inference margins, and vertical integration shape the future of labs like OpenAI and Anthropic.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 15.2% of the talking time here. How this is scored →

The hosts as informed peer 3.3 Guest teaching 5.0 Guest disagreement 1.1 The hosts pushing back 1.6
05100:0010:0020:000:56–5:42 · The hosts as informed peer 2/10 Explaining Cournot Equilibrium in AI Frontier Competition Packy delivers an extensive breakdown of the Cournot equilibrium applied to AI frontier labs, detailing the distinction between training capex depreciation and inference manufacturing margins. The hosts mostly chime in with chat reactions and humorous titling context.5:43–9:55 · The hosts as informed peer 6/10 Frontier Pricing Dynamics and Practical Limits of Base Models John Coogan directly pushes back against Packy's tax code example, arguing that large context windows excel at parsing complex documents. Packy counters by explaining the legal nuances and regulatory discretion that base models cannot capture without specialized fine-tuning.9:56–12:48 · The hosts as informed peer 6/10 Market Evolution from Cournot to Bertrand Competition Jordi demonstrates domain knowledge by analyzing Anthropic's API vulnerability versus application-layer leverage. Packy expands on this by explaining the theoretical transition from Cournot oligopoly to Bertrand price competition once frontier intelligence commoditizes.12:48–14:52 · The hosts as informed peer 2/10 NeoClouds, Vertical Integration, and OpenAI's Strategic Moves After brief banter regarding a hypothetical Malibu NeoCloud, Packy details OpenAI's recent moves toward vertical integration, citing the Cerebras partnership, OpenClaw routing, and enterprise forward-deployed engineers.14:53–17:53 · The hosts as informed peer 4/10 Dario Amodei and Dwarkesh Patel on AI Scaling Economics In the featured podcast clip, Dario Amodei answers Dwarkesh Patel and John Coogan's question about current lack of profits by clarifying that positive per-model gross margins are masked by exponential scale-up capex for next-generation training.17:53–21:30 · The hosts as informed peer 1/10 Sponsor Break: Turbopuffer This segment is dedicated to a sponsor read and playing a clip from A Beautiful Mind to illustrate Nash equilibrium and game theory.21:31–22:48 · The hosts as informed peer 2/10 Stream Chat Reactions and School Movie Day Memories The episode wraps up with casual chat interaction, nostalgic banter about movie days in school, and a final summary of game theory's role in the long-term AI market structure.0:56–5:42 · Guest teaching 7/10 Explaining Cournot Equilibrium in AI Frontier Competition Packy delivers an extensive breakdown of the Cournot equilibrium applied to AI frontier labs, detailing the distinction between training capex depreciation and inference manufacturing margins. The hosts mostly chime in with chat reactions and humorous titling context.5:43–9:55 · Guest teaching 5/10 Frontier Pricing Dynamics and Practical Limits of Base Models John Coogan directly pushes back against Packy's tax code example, arguing that large context windows excel at parsing complex documents. Packy counters by explaining the legal nuances and regulatory discretion that base models cannot capture without specialized fine-tuning.9:56–12:48 · Guest teaching 5/10 Market Evolution from Cournot to Bertrand Competition Jordi demonstrates domain knowledge by analyzing Anthropic's API vulnerability versus application-layer leverage. Packy expands on this by explaining the theoretical transition from Cournot oligopoly to Bertrand price competition once frontier intelligence commoditizes.12:48–14:52 · Guest teaching 5/10 NeoClouds, Vertical Integration, and OpenAI's Strategic Moves After brief banter regarding a hypothetical Malibu NeoCloud, Packy details OpenAI's recent moves toward vertical integration, citing the Cerebras partnership, OpenClaw routing, and enterprise forward-deployed engineers.14:53–17:53 · Guest teaching 8/10 Dario Amodei and Dwarkesh Patel on AI Scaling Economics In the featured podcast clip, Dario Amodei answers Dwarkesh Patel and John Coogan's question about current lack of profits by clarifying that positive per-model gross margins are masked by exponential scale-up capex for next-generation training.17:53–21:30 · Guest teaching 2/10 Sponsor Break: Turbopuffer This segment is dedicated to a sponsor read and playing a clip from A Beautiful Mind to illustrate Nash equilibrium and game theory.21:31–22:48 · Guest teaching 3/10 Stream Chat Reactions and School Movie Day Memories The episode wraps up with casual chat interaction, nostalgic banter about movie days in school, and a final summary of game theory's role in the long-term AI market structure.0:56–5:42 · Guest disagreement 1/10 Explaining Cournot Equilibrium in AI Frontier Competition Packy delivers an extensive breakdown of the Cournot equilibrium applied to AI frontier labs, detailing the distinction between training capex depreciation and inference manufacturing margins. The hosts mostly chime in with chat reactions and humorous titling context.5:43–9:55 · Guest disagreement 4/10 Frontier Pricing Dynamics and Practical Limits of Base Models John Coogan directly pushes back against Packy's tax code example, arguing that large context windows excel at parsing complex documents. Packy counters by explaining the legal nuances and regulatory discretion that base models cannot capture without specialized fine-tuning.9:56–12:48 · Guest disagreement 1/10 Market Evolution from Cournot to Bertrand Competition Jordi demonstrates domain knowledge by analyzing Anthropic's API vulnerability versus application-layer leverage. Packy expands on this by explaining the theoretical transition from Cournot oligopoly to Bertrand price competition once frontier intelligence commoditizes.12:48–14:52 · Guest disagreement 0/10 NeoClouds, Vertical Integration, and OpenAI's Strategic Moves After brief banter regarding a hypothetical Malibu NeoCloud, Packy details OpenAI's recent moves toward vertical integration, citing the Cerebras partnership, OpenClaw routing, and enterprise forward-deployed engineers.14:53–17:53 · Guest disagreement 2/10 Dario Amodei and Dwarkesh Patel on AI Scaling Economics In the featured podcast clip, Dario Amodei answers Dwarkesh Patel and John Coogan's question about current lack of profits by clarifying that positive per-model gross margins are masked by exponential scale-up capex for next-generation training.17:53–21:30 · Guest disagreement 0/10 Sponsor Break: Turbopuffer This segment is dedicated to a sponsor read and playing a clip from A Beautiful Mind to illustrate Nash equilibrium and game theory.21:31–22:48 · Guest disagreement 0/10 Stream Chat Reactions and School Movie Day Memories The episode wraps up with casual chat interaction, nostalgic banter about movie days in school, and a final summary of game theory's role in the long-term AI market structure.0:56–5:42 · The hosts pushing back 0/10 Explaining Cournot Equilibrium in AI Frontier Competition Packy delivers an extensive breakdown of the Cournot equilibrium applied to AI frontier labs, detailing the distinction between training capex depreciation and inference manufacturing margins. The hosts mostly chime in with chat reactions and humorous titling context.5:43–9:55 · The hosts pushing back 6/10 Frontier Pricing Dynamics and Practical Limits of Base Models John Coogan directly pushes back against Packy's tax code example, arguing that large context windows excel at parsing complex documents. Packy counters by explaining the legal nuances and regulatory discretion that base models cannot capture without specialized fine-tuning.9:56–12:48 · The hosts pushing back 1/10 Market Evolution from Cournot to Bertrand Competition Jordi demonstrates domain knowledge by analyzing Anthropic's API vulnerability versus application-layer leverage. Packy expands on this by explaining the theoretical transition from Cournot oligopoly to Bertrand price competition once frontier intelligence commoditizes.12:48–14:52 · The hosts pushing back 0/10 NeoClouds, Vertical Integration, and OpenAI's Strategic Moves After brief banter regarding a hypothetical Malibu NeoCloud, Packy details OpenAI's recent moves toward vertical integration, citing the Cerebras partnership, OpenClaw routing, and enterprise forward-deployed engineers.14:53–17:53 · The hosts pushing back 4/10 Dario Amodei and Dwarkesh Patel on AI Scaling Economics In the featured podcast clip, Dario Amodei answers Dwarkesh Patel and John Coogan's question about current lack of profits by clarifying that positive per-model gross margins are masked by exponential scale-up capex for next-generation training.17:53–21:30 · The hosts pushing back 0/10 Sponsor Break: Turbopuffer This segment is dedicated to a sponsor read and playing a clip from A Beautiful Mind to illustrate Nash equilibrium and game theory.21:31–22:48 · The hosts pushing back 0/10 Stream Chat Reactions and School Movie Day Memories The episode wraps up with casual chat interaction, nostalgic banter about movie days in school, and a final summary of game theory's role in the long-term AI market structure.

speaking balance: gold is the hosts, purple is the guest (3 minute bins)

0:00 · the hosts 24.9% · guest 75.1%0:00 · the hosts 24.9% · guest 75.1%3:00 · the hosts 0.5% · guest 99.5%3:00 · the hosts 0.5% · guest 99.5%6:00 · the hosts 21.2% · guest 78.8%6:00 · the hosts 21.2% · guest 78.8%9:00 · the hosts 27.7% · guest 72.3%9:00 · the hosts 27.7% · guest 72.3%12:00 · the hosts 14.8% · guest 85.2%12:00 · the hosts 14.8% · guest 85.2%15:00 · the hosts 6.1% · guest 93.9%15:00 · the hosts 6.1% · guest 93.9%18:00 · the hosts 13.9% · guest 86.1%18:00 · the hosts 13.9% · guest 86.1%21:00 · the hosts 9.9% · guest 90.1%21:00 · the hosts 9.9% · guest 90.1%
Sharpest disagreement ▶ 8:48 Packy challenges John's context window assumption

Packy rejects John's assertion that feeding tax code into a prompt solves complex financial edge cases, arguing that legal enforcement discretion requires deeper human nuance.

 Hardest push from the hosts ▶ 8:36 John Coogan challenges Packy's tax code example

John refuses the premise that models fail at complex tax regulations, pointing out that extracting specific facts from large documents is precisely what LLMs do best.

 Biggest teaching moment ▶ 16:31 Dario Amodei breaks down AI lab P&L realities

Dario clearly educates the interviewer by showing how a model producing $2 billion net profit is overshadowed on the balance sheet by spending $10 billion on the next generation.

 The host holds their own ▶ 10:32 Jordi articulates Anthropic's product vs API risk

Jordi demonstrates clear strategic expertise by outlining why an API-only business model leaves labs trapped on a model-training hamster wheel unless protected by application-level UX.
the scores for every segment, with the reasoning behind each
ChapterTopicThe hosts as informed peerGuest teachingGuest disagreementThe hosts pushing backWhy
Explaining Cournot Equilibrium in AI Frontier Competition 2710 Packy delivers an extensive breakdown of the Cournot equilibrium applied to AI frontier labs, detailing the distinction between training capex depreciation and inference manufacturing margins. The hosts mostly chime in with chat reactions and humorous titling context.
Frontier Pricing Dynamics and Practical Limits of Base Models 6546 John Coogan directly pushes back against Packy's tax code example, arguing that large context windows excel at parsing complex documents. Packy counters by explaining the legal nuances and regulatory discretion that base models cannot capture without specialized fine-tuning.
Market Evolution from Cournot to Bertrand Competition 6511 Jordi demonstrates domain knowledge by analyzing Anthropic's API vulnerability versus application-layer leverage. Packy expands on this by explaining the theoretical transition from Cournot oligopoly to Bertrand price competition once frontier intelligence commoditizes.
NeoClouds, Vertical Integration, and OpenAI's Strategic Moves 2500 After brief banter regarding a hypothetical Malibu NeoCloud, Packy details OpenAI's recent moves toward vertical integration, citing the Cerebras partnership, OpenClaw routing, and enterprise forward-deployed engineers.
Dario Amodei and Dwarkesh Patel on AI Scaling Economics 4824 In the featured podcast clip, Dario Amodei answers Dwarkesh Patel and John Coogan's question about current lack of profits by clarifying that positive per-model gross margins are masked by exponential scale-up capex for next-generation training.
Sponsor Break: Turbopuffer 1200 This segment is dedicated to a sponsor read and playing a clip from A Beautiful Mind to illustrate Nash equilibrium and game theory.
Stream Chat Reactions and School Movie Day Memories 2300 The episode wraps up with casual chat interaction, nostalgic banter about movie days in school, and a final summary of game theory's role in the long-term AI market structure.

Statements from this episode (11)

Opinion
McCormick: AI labs obsess over competitor compute despite public denials
“And this is really, really relevant to the AI lab discussion because you can tell that even though all of the leaders of the AI lab say, I don't think about the competition. I don't talk about the competition. I'll use general terms. They're all obsessed with …”
Packy McCormick Feb 18, 2026 ▶ 1:38
Opinion
McCormick: OpenAI and Anthropic are the most unprofitable companies in history
“These are the most unprofitable companies in human history, I think.”
Packy McCormick Feb 18, 2026 ▶ 3:11
Assertion Supported
McCormick: OpenAI recouped all GPT-4 training costs and more
“GPT four is the really instructive is the really instructive example because I believe that model costs like a hundred million dollars to train, and it was really expensive at the time, but then very quickly they were on a multi-billion dollar run rate, and it…”
Packy McCormick Feb 18, 2026 ▶ 4:13
Opinion
McCormick: OpenAI and Anthropic inference margins are healthy
“Well, it's way, way cheaper than what you pay to Anthropic or OpenAI, so they must have good margins, and everyone sort of agrees at this point that inference margins are in fact healthy.”
Packy McCormick Feb 18, 2026 ▶ 5:25
Assertion Supported
McCormick: OpenAI signed a deal with Cerebras for fast inference
“OpenAI just did the Cerebrus deal, there's Claude Fast, and there's a whole bunch of different modes that will deliver faster inference”
Packy McCormick Feb 18, 2026 ▶ 6:03
Prediction Not checkable as stated
McCormick: Models mastering complex tax and legal nuances is six months away
“This is like a six month away thing.”
Packy McCormick Feb 18, 2026 ▶ 9:43
Insight
Hays: Product applications give AI companies more leverage than pure APIs
“Having a product, not just an API business, gives you leverage, because at some point, the models are smart enough where you don't need to train them, you don't need to train a model that is four percent better, because people are still coming to your applicat…”
Jordi Hays Feb 18, 2026 ▶ 10:32
Prediction Not checkable as stated
McCormick: Frontier AI will settle into an oligopoly among 3-4 labs
“A lot of the VC firms are getting in multiple companies because they don't think it's going to be winner-take-all anymore. They think it's going to be much more oligopolistic for the long term, and there will be competition between the major three or four labs…”
Packy McCormick Feb 18, 2026 ▶ 12:16
Insight
Amodei: AI market with three rational firms will not collapse to zero margins
“The point is it doesn't equilibrate to perfect competition with zero margins. If there's like three firms, if there's three firms in the economy, All are kind of independently behaving, behaving rationally. It doesn't equilibrate to zero.”
Dario Amodei Feb 18, 2026 ▶ 16:00
Insight
Amodei: AI models make money individually while companies lose money scaling
“Each model makes money, but the company loses money.”
Dario Amodei Feb 18, 2026 ▶ 17:30
Prediction Not checkable as stated
McCormick: AI labs will become profitable hyperscalers like cloud
“It seems like the labs will Turn into sort of new hyperscalers. There will be you know, increased competition, but still very, very good businesses a la cloud.”
Packy McCormick Feb 18, 2026 ▶ 22:38
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 500 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more