How DeepStack settles a question · Part 05
Do frontier labs sell a commodity?
What does a frontier lab sell: a commodity, or a premium?
From Part 05: OpenAI and Anthropic Are Not the Same Business. 5 voices on 2 sides, attributed and dated; the DeepStack Call with its history; 5 dated tests that settle it; the results as they come in.
The mechanism
Hyperscalers, lenders and investors have priced the two labs as one AI bet, and the revenue of the two behaves differently under pressure. OpenAI cut its API price by 80 percent as cheaper open and Chinese models took the routing, and lost $20.9 billion in 2025; Anthropic grew 142 percent quarter on quarter in the second quarter of 2026 against OpenAI's 18 percent, on enterprise coding that still monetizes, with $559 million of preliminary adjusted operating income for the June quarter. One revenue line holds when the token gets cheaper; the other has to be bought with price cuts. That is why the Call splits the trade rather than taking a side: what a lab sells depends on whether its margin survives the cluster's useful life, which only audited disclosure can show.
Approved by Marco Cauduro on October 10, 2026.
The DeepStack callSplit
Stop pricing OpenAI and Anthropic as one bet. One revenue line holds under pressure; the other has to be bought.
Our Call takes neither side alone.
Call history
- v2Split the tradeConviction Moderate (3/5)SynthesisSide named by the owner.
- v1Split the tradeConviction Moderate (3/5)First call, as published with the story. The conviction, the actions and the flips were set down for the site on October 7, 2026.
4 voices
OpenAI: commoditized
Michael Spencer predicts Anthropic will achieve AI supremacy status in 2027 and by 2030 will have no comparable peer competitor, not Google, OpenAI, DeepSeek, or Alibaba.
Open-source Chinese models are disrupting OpenAI's revenue because routing and code wrappers increasingly favor them over OpenAI's models due to cheaper token costs, forcing OpenAI to cut API prices, including an 80% cut to GPT-5.6 Lune, which…
Chinese open-weight models climbed from under 2% of token traffic in late 2024 to a majority of usage among top models in 2026, driven by the revealed preference for models running on infrastructure the user controls.
In June 2026, the Commerce Department suspended access to Anthropic's Fable 5 and Mythos 5 models for all non-US persons, invoking export-control authority, with practical effect of a global suspension that lasted under three weeks before…
4 voices
Anthropic: not yet
Training runs account for only 10 to 23 percent of AI labs' total compute costs, with the remainder tied to tacit knowledge in data curation, infrastructure, and learning from failed experiments.
The demand curve for AI capabilities is heterogeneous: commodity intelligence will become price-competitive across many applications, while convex-payoff domains like finance and frontier research can sustain premium pricing for marginal capability…
Anthropic's ARR grew from approximately $700 million in August 2024 to its current level, a roughly 93x increase in about two years.
Cognition's Devin achieved a $1 billion annualized revenue run rate and a $40 billion valuation in under two years, making it exceptional among AI startups for this pace of revenue growth.
6 voices
Voices that cross sides
As Devin and similar tools scale, they drive increased token usage of Anthropic's models, creating a compounding benefit to Anthropic.
Europe's dream of sovereign AI leadership is fading because Mistral, despite raising 3.5 billion dollars at a 21 billion euro valuation, has been unable to compete with the United States or China in open-source AI models.
Wyndo cites Artificial Analysis benchmark results as evidence of rapid open-weight model improvement: Kimi K3 scored 57 on its Intelligence Index (13 points above Kimi K2.6, per the July 17 evaluation), and DeepSeek V4 Flash 0731 scored 10 points…
The traditional split of using a strongest frontier model (Opus) for reasoning/judgment and a faster cheaper model (Sonnet) for execution has become less clearly justified because the capability gap between 'thinking' and 'executing' models has…
Because OpenAI and Anthropic pay their bills with venture funding of which Big Tech supplies roughly 40%, revenue Big Tech books from them is partly recycled from itself and not real growth.
If Big Tech depends heavily on OpenAI and Anthropic for revenue growth, that growth is unsustainable because both customers are massively unprofitable and may not survive.
Where they cross
Who answers whom.
The specific points where one voice meets another: the same evidence read two ways, or the same mechanism with a different sign.
Ed Elson qualifies Michael Spencer, on a condition
Michael Spencer contradicts Ed Elson, on premise
Michael Spencer qualifies Wyndo, on a condition
Alternative explanations
Three other readings.
What would have to be true for the Call to be wrong, as the other side argues it.
- Neither lab keeps a premium: Chinese open-weight models went from under 2 percent of token traffic in late 2024 to a majority of usage among the top models in 2026 (the case Massif Capital makes).
- The premium holds where the payoff is convex: commodity intelligence turns price-competitive in many applications, while finance and frontier research sustain premium pricing (the case Christian Catalini makes).
- Neither revenue line is what it seems: the labs pay their bills with venture funding of which Big Tech supplies roughly 40 percent, so the revenue Big Tech books from them is partly recycled (the case Ed Elson makes).
What settles it
The dated tests.
The tests the story set, each with its specification where the owner has written one and the piece of the argument it verifies; on the Docket, resolved on the Results page.
- Next prospectus
An audited listing reveals margins, customer concentration and investor-linked revenue; clean disclosure would support separation, weak economics would reunite the risks.
Part 05: OpenAI and Anthropic Are Not the Same Business Pendingp05-t01Open questionChecks the defensible-margin claim
Specification pending
- Next disclosures
Useful-life assumptions show whether first-year GPU margins survive across the cluster; Romero’s accounting would gain force if they do not.
Part 05: OpenAI and Anthropic Are Not the Same Business Pendingp05-t02Open questionChecks the cluster-margin claim
Specification pending
- DeepSeek repricing
A quiet reversal would weaken Erkan’s demand evidence; sustained higher prices would support the view that demand remains structurally strong.
Part 05: OpenAI and Anthropic Are Not the Same Business Pendingp05-t03Open questionChecks the demand-evidence claim
Specification pending
Return on invested capital should be visible by Romero’s deadline; failure would turn today’s extraordinary economics into a first-year illusion.
Part 05: OpenAI and Anthropic Are Not the Same Business Pendingp05-t04Open questionChecks the return-on-capital claim
Specification pending
- A second revocation
Another model suspension would turn Massif’s signal into procurement policy, showing that open weights are changing buying behavior.
Part 05: OpenAI and Anthropic Are Not the Same Business Pendingp05-t05Open questionChecks the open-weights procurement claim
Specification pending
What changed
The record of Part 05.
The Part’s publication, every change to the Call, every result that came in and the next dated test, oldest first.
- PublishedPart 05 published; the Call: Split the trade.
- Call updatedThe Call: Split the trade. Stop pricing OpenAI and Anthropic as one bet. One revenue line holds under pressure; the other has to be bought. Our Call takes neither side alone. by Marco Cauduro
- Next testNext: Jan 3, 2028, Return on invested capital should be visible by Romero’s deadline; failure would turn today’s extraordinary economics into a first-year illusion. Open question
Sources (6)
- AI Supremacy - Q2 lab revenue comparison
- Fiscal.ai - Massif Capital Q2 letter
- a16z News - Open versus closed AI economics
- Capitalist Letters - Frontier compute economics
- Hypertech Invest - AI selloff analysis
- Financial Modeling Prep - Palantir Q2 call
Market data are as of the dates cited. Voices are paraphrased from public writing and attributed by name.
Analysis and opinion, not investment advice. Voices are paraphrased from public writing and attributed by name; DeepStack shows no score, ranking or accuracy for any person. Every question, one page each.