Palantir chief executive Alex Karp told its third-quarter earnings call on 2 November that customers are tired of “software companies that are reselling tokens and it counts as revenue”. Minutes later he called Anthropic’s chief executive, Dario Amodei, “one of the most important business people of my generation”.
The two companies report their main cost lines differently. By our arithmetic, Anthropic, which sells usage “metered in tokens”, spent 58 cents of each revenue dollar on compute and infrastructure in the third quarter. Palantir’s cost of revenue was 17 cents. The two measures are not like-for-like.
The webcast · Palantir Q3 2026 earnings call
Palantir’s own stream of its third-quarter 2026 earnings webcast, 2 November 2026. Sankar’s remarks on the model makers start at 9:53; Karp’s remarks on token resellers are at 35:35 and on Anthropic at 37:52.
An evergreen Business Engineer explainer on gross margin: what an AI company keeps from each dollar. It teaches the general idea only and does not discuss Palantir or Anthropic.
The key insight: As we read it, Palantir drew its line on the call between token sellers and the layer that runs work: Karp faulted firms whose resold tokens count as revenue, while Sankar said the labs have the unit of the enterprise wrong.
The first analyst question asked how Palantir’s AI boot camps turn visitors into customers. Near the end of a long answer, at [35:35], Karp said: “The other stuff is not working”.
At [35:40] he referred to “a very strong sense in the market” about “these other approaches”, then listed them. One phrase in that list is unclear in our transcription, and we do not quote it. The rest reads: “lightweight companies reselling tokens”, and, at [35:54], “software companies that are reselling tokens and it counts as revenue”.
At [35:59] he added: “People are just tired of it”. At [36:02]: “the taste of the market is shifting”. Karp did not name any company, and he did not describe Anthropic as one of them.
Cents per $1 of quarterly revenue: Anthropic’s compute and infrastructure expense and Palantir’s cost of revenue, first quarter 2025 to third quarter 2026. Sources: Anthropic’s amended S-1 (2 November 2026); Palantir’s SEC filings and Q3 2026 release. Our arithmetic; Palantir Q4 2025 is the full year minus nine months. Not like-for-like: Anthropic’s compute line includes research and model training.Animated: Anthropic’s compute and infrastructure expense in cents per dollar of quarterly revenue, first quarter of 2025 to third quarter of 2026, our arithmetic from Anthropic’s amended S-1; Palantir’s third-quarter cost of revenue (17 cents) from its Q3 2026 release. Not like-for-like: Anthropic’s compute line includes research and model training. Values are computed from filed figures, not read from a chart. Chart: FourWeekMBA.
What Sankar Said About the Model Makers
Chief technology officer Shyam Sankar set out the argument in his prepared remarks. At [09:53] he said: “Palantir is obsessed with turning tokens into economic value”. At [10:09]: “On one end, you have the model makers who can’t conceive of solving any problem except through model improvement. So they aren’t even trying”.
At [10:38] he said “the conceit that the labs have wrong is the belief that the atomic unit of the enterprise is the individual knowledge worker”. At [10:48]: “The fundamental unit of the enterprise is the workflow”. He made one exception at [10:56]: “For some things, like coding, the knowledge worker is the workflow”.
His examples included agent swarms. At [12:32] he described “An agent swarm per wellhead in the Permian Basin to maximize production”, and at [12:36] “An agent swarm for every customer of our customer to drive growth”.
What Karp Said About Anthropic
Asked by an analyst from D.A. Davidson about chief executives who are not yet convinced, Karp said at [37:52]: “Dario is someone I get along very well with”. He continued: “I think he’s one of the most important business people of my generation, if not multiple generations”.
At [38:23] he drew the line between the two companies: “what they’re doing is increasing the size of the market. What we’re doing is servicing the market that’s gotten real”. His other remarks on Anthropic, including the word “symbiotic”, are in our piece on the AIG workflow.
His shareholder letter, published the same day, makes the written case: “In the best case, these models on their own generate limited value”.
What the Filings Show
Anthropic no longer reports a gross margin. Its S-1 says that during 2025 it changed its presentation so that “cost of revenue, gross profit (loss), research and development, sales and marketing, and general and administrative are no longer presented as separate line items”.
Its largest line is compute and infrastructure, which the S-1 says supports “the full range of workloads for our models, including research, model development, serving customers, and internal usage of Claude”. In the third quarter that line was $9.863 billion against revenue of $16.956 billion, according to the amended S-1.
By our arithmetic that is 58 cents per revenue dollar, up from 48 cents in the second quarter and down from $1.48 a year earlier. Palantir’s cost of revenue was $383 million on revenue of $2.313 billion, or 17 cents, a gross margin of 83%. By the same arithmetic on its filings, Palantir’s figure was 13 cents in the first quarter of 2026 and 18 cents a year earlier.
As we read it, the argument is about which layer of the stack keeps the revenue: Karp described Anthropic as growing the market and Palantir as servicing the part that has become real.
As we read it, the filings show why the layer matters to margins: Anthropic’s main cost line took 58 cents of each revenue dollar in the third quarter, while Palantir’s cost of revenue took 17 cents, on measures that are not like-for-like.
As we read it, Karp’s criticism was aimed at resellers rather than at Anthropic, which he praised in the same call.
Alex Karp, Palantir CEO, on the Q3 2026 earnings call
“software companies that are reselling tokens and it counts as revenue. People are just tired of it”
Three Implications
FOR INVESTORS TRACKING PALANTIR Its chief executive set Palantir apart from firms that resell tokens; its cost of revenue was 17 cents per revenue dollar in the quarter, by our arithmetic.
FOR SOFTWARE COMPANIES BUILT ON MODEL APIS Karp said customers are “just tired of it” when resold tokens count as revenue; he named no company.
FOR ANTHROPIC WATCHERS Anthropic sells usage “metered in tokens” and no longer reports a gross margin; its compute line took 58 cents per revenue dollar in the third quarter, by our arithmetic.
The framework argues: “It is simultaneously the unit of cognition the model produces, the unit of compute the data center serves, the unit of price the lab charges, and the unit of value the enterprise extracts.”
The framework is describing the token. As we read it, Sankar’s call remarks contest the last part of that sentence: he said the unit that matters to the enterprise is the workflow, and Karp set Palantir against firms that resell the tokens. The framework was published in June 2026 and does not discuss Palantir’s results.
Karp did not say which companies he meant by token resellers or how their revenue is booked. Anthropic’s compute line includes research and model training, which a cost of revenue would not, so the 58-cent and 17-cent figures measure different things. They show how much of each revenue dollar goes to the main cost line on each side, not which business is more profitable.
The quotations come from our machine transcription; we found no official transcript of the call. We did not contact Palantir or Anthropic.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growth story. This week it started being priced as a credit story.
Karp told Palantir’s call that customers are tired of “software companies that are reselling tokens and it counts as revenue”, while praising Anthropic’s chief executive. By our arithmetic, Anthropic spent 58 cents of each third-quarter revenue dollar on compute and infrastructure; Palantir’s cost of revenue was 17 cents.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Quotations attributed to Alex Karp and Shyam Sankar come from our own machine transcription of the audio of Palantir’s third-quarter 2026 earnings webcast, streamed on Palantir’s YouTube channel on 2 November; it is not an official transcript, and timestamps refer to that video. Anthropic’s figures and definitions come from its S-1 of 26 October 2026 and amended S-1 of 2 November; Palantir’s from its earnings release of 2 November (8-K Exhibit 99.1) and its SEC filings. Cents-per-dollar figures are our arithmetic and are not like-for-like. We did not contact Palantir or Anthropic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
