On Monday August 3, 2026, Palantir reported second-quarter revenue of $1.94 billion, up 93 per cent year-over-year, with US commercial revenue up 149 per cent and net income of $1.07 billion. Full-year 2026 guidance was raised to $8.15–$8.16 billion. In the same shareholder letter, CEO Alex Karp — PhD in social theory, defence-contractor CEO, co-author of The Technological Republic — accused OpenAI, Anthropic and other frontier AI labs of having "Marxist overtones and undertones" and of intending "to capture the means of production of their purported partners."

A quarter that big is what most CEOs celebrate with a victory lap and a modest sentence about strong momentum. Karp celebrated by reaching for Marx and not missing.

What the shareholder letter actually says

The sentence people are quoting, via TechCrunch's writeup and Yellow's roll-up of the letter, runs in two clauses:

"There are Marxist overtones and undertones to our business."

"Others, including many of those building large language models, intend, knowingly or otherwise, to capture the means of production of their purported partners."

The theory Karp is proposing — and it is a theory, dressed in the vocabulary of a PhD student pushing back on his committee — is that enterprises today pay a subscription fee to a foundation-model vendor, feed it their proprietary workflows and intellectual property, and by that act fund the model that will one day replace their own workforce. The customer is the fuel; the model is the fire; the fire eventually eats the customer. Reframed in Marxist terms: the workers are alienated from the tools they produce, because the tools now produce themselves.

That is a real argument. It is the same argument Anthropic's own economic-index papers keep landing on — that generative AI has an unusually compressed distance between the input signal (enterprise data) and the output substitute (an agent that does the enterprise's job). What is unusual is a defence-contractor CEO saying it out loud from a shareholder letter, and using the vocabulary of nineteenth-century political economy to do so.

The earnings-call line the trade press keeps quoting

On the earnings call, via the Yahoo Finance readout of TechCrunch's account, Karp asked whether enterprises were going to "buy into a future" where their work helps their "adversaries win, and everybody who does win is a small, tiny group of people living in a tiny place that somehow believe because they eat vegetables and they don't support war fighters that they deserve to have the total means of production of this country?"

That is one sentence. It contains: an adversary framing, a coastal-elite geography ("a tiny place"), a dietary slur ("eat vegetables"), a defence-industrial-base litmus test ("don't support war fighters"), and the Marxist term of art ("means of production") repeated for the second time in a single day. Karp is a philosophy PhD who does not use these words by accident. He is reading a country in which AI-lab San Francisco is one coast and Palantir Denver is another, and asking whose model of value creation the enterprise buyer wants to fund.

The "vegetables" line will be the one to survive on X. The "means of production" line is the one his shareholders paid for.

The financial context makes the accusation possible

Palantir's Q2 was, by InvestingLive's summary, the strongest quarter it has ever reported. The numbers Karp opened the letter with:

• Revenue: $1.94 billion, up 93 per cent year-over-year
• US commercial revenue: $764 million, up 149 per cent
• US government revenue: $809 million, up 90 per cent
• GAAP net income: $1.07 billion — a 55 per cent margin
• Adjusted free cash flow: $1.2 billion — a 63 per cent margin
• Rule of 40: 155 per cent — a number that reads like a typo and is not

Karp described the quarter as "otherworldly." Full-year US-commercial guidance was raised to more than $3.4 billion, up 134 per cent. Adjusted operating income for the year is now guided to $4.9 billion. These are the numbers of a company that has stopped worrying about whether it is winning.

A CEO who has just handed shareholders a 93-per-cent revenue quarter can afford to be more interesting in his shareholder letter than the average earnings-day boilerplate. Karp used the space that in other companies would have contained the phrase "strong tailwinds in the enterprise AI market" to accuse the enterprise-AI market of ideological hypocrisy. It reads different when the person doing it is running a 155-per-cent Rule of 40.

The Palantir pitch, decoded

Underneath the vocabulary is a sales pitch, and the sales pitch is the reason the earnings call could sustain the vocabulary. Palantir's product story for the last three years has been AI sovereignty: the customer's models, prompts and workflows run on the customer's infrastructure, on model weights the customer can swap, with data the customer keeps. Foundry and AIP are model-agnostic on purpose. The upgrade path is choose a different foundation model next year, not migrate off Palantir.

The Marxist metaphor is Karp translating that architecture into ideology. In his account, an OpenAI or Anthropic contract is Marxist because the customer's labour (their data, their prompt engineering, their edge-case corrections) accrues to the vendor's model, and the vendor owns the resulting improvement. A Palantir contract is not Marxist because the labour accrues to the customer's own environment. Whether that description survives a serious look at how enterprise Foundry deployments actually work is a separate question; as a rhetorical stroke it is closed. He has turned lock-in into class struggle, and the buyer into a proletarian.

The trick works because a fraction of Palantir's enterprise-customer base is already wary of foundation-model vendors after this month's news. Anthropic's July 30 cyber-eval disclosure — three real breaches out of 141,006 runs, one involving a real Python package fifteen real machines executed — is fresh in every CISO's head. OpenAI's July 21 Hugging Face incident is fresher. Karp is arriving at the earnings call, in the same week those events are still being written up, with a pitch that says the labs are Marxist and unreliable; we are neither. It is not a bad week to be selling that message.

Where the argument does not hold

Karp's letter is quotable and, in parts, defensible. It also has three weak seams a critical reader will find quickly.

One: Palantir is not neutral about its own model captures. The Rule-of-40 number that reads like a typo — 155 per cent — comes from a business whose net-revenue-retention rates and Total Contract Value expansion figures (US commercial TCV up 153 per cent to $2.1 billion) are also, by any honest accounting, forms of lock-in. Foundry is not model-locked; Foundry customers, in aggregate, are Foundry-locked. Calling frontier-lab arrangements Marxist while running a 155-per-cent Rule of 40 is a specific rhetorical move. It is not the same as being neutral.

Two: the "means of production" line elides the training-data question. Enterprise foundation-model contracts today, at OpenAI-scale and Anthropic-scale, explicitly do not train on customer data by default. That protection is imperfect, contested, and priced into the tier — but it is a term of art the metaphor pretends does not exist. If OpenAI is Marxist for capturing enterprise labour into a shared model, the sentence has to argue that despite the no-training clauses, the capture happens anyway (through prompt structure, through evaluation feedback, through the models people learn to write for). Karp does not make that argument; he leaves the metaphor to do the work.

Three: the moral geography is a caricature. "A small, tiny group of people living in a tiny place that somehow believe because they eat vegetables and they don't support war fighters" is San Francisco written into three modifiers. It is fun. It is also the argument-by-vibes version of a real question — whether the values of a small demographic of AI-lab employees are the values American enterprises should route their capabilities through. A defence-contractor CEO is entitled to raise the question. He is not entitled to have the caricature substitute for the argument.

What to watch

1. Whether Sam Altman or Dario Amodei responds. Karp named OpenAI and Anthropic in a public letter. Both CEOs have a well-developed habit of engaging with critics of their pricing and safety story (Amodei took twelve days to answer the open-weights case; Altman's July 28 podcast run was one long response to the Pacing the Frontier letter). If either engages the Marxist framing directly, it becomes a two-sided debate. If neither does — the more likely outcome — it becomes a Karp monologue that shapes the enterprise-buyer conversation on its own.
2. Whether the "AI sovereignty" pitch shows up in Q3 procurement conversations. Karp's rhetoric is downstream of a sales motion that has to close deals. If Foundry AIP starts appearing in Q3 RFPs specifically as an alternative to OpenAI or Anthropic because of the sovereignty framing, the letter did commercial work, not just narrative work. Bloomberg and The Information enterprise-tech desks will surface this within four to six weeks.
3. How the Karp-Altman split plays into the AI-industry lobbying map. Palantir has been on the war-fighter side of the AI-industry lobbying story all year; OpenAI and Anthropic both raised their lobbying spend to record levels in Q2. Karp is now openly framing that split as ideological rather than sectoral. Watch whether that framing gets picked up by Republican senators (Cruz, Moreno, Hawley) or defence-appropriations staff over the next month.
4. Whether the "otherworldly" quarter looks like a peak. Palantir's Q2 was the strongest in its history. Q3 has to compare against a base that just moved. If the deceleration narrative surfaces in October's earnings preview cycle, the Marxist letter will read differently in retrospect — as an offensive move from a company that could still afford to make one. If Q3 comes in above Q2, the letter reads as the opening of a two-year positioning war for the enterprise-AI wallet.

The load-bearing sentence in Karp's letter is not the Marxist one. It is this: "Others, including many of those building large language models, intend, knowingly or otherwise, to capture the means of production of their purported partners." A defence-contractor CEO used a shareholder letter to accuse his enterprise-software rivals of engineering a class relationship with their customers. He did it in the same document where he raised full-year guidance by half a billion dollars.

The AI industry's most-quoted book this week was written by Karl Marx, and the most-quoted CEO reading it lives in Denver.