Palantir's Q2 2026 Earnings and Karp's AI Labs Critique

Palantir's Q2 2026 print set records across nearly every line: total revenue of $1.935 billion, up 93% year-over-year, the company's highest-ever growth rate ^[3]. U.S. commercial revenue - the segment investors watch most closely for proof of enterprise adoption beyond government contracts - grew 149% to $764 million, while total U.S. revenue hit $1.573 billion, now more than 81% of the company's business ^[4]. Profitability kept pace: $1.062 billion in GAAP net income at a 55% margin, $1.22 billion in adjusted free cash flow, and adjusted EPS of $0.41 against a $0.35 consensus - the first quarter Palantir cleared $1 billion simultaneously across net income, adjusted FCF, and adjusted operating income ^[3][5]. Deal velocity backed up the growth story: 220 deals worth at least $1 million, including 73 above $10 million ^[5]. Palantir raised its full-year revenue guidance to $8.15-$8.158 billion (82% growth) and its U.S. commercial growth guidance to at least 134% ^[3]. Shares responded with a 15-20% single-session rally ^[4][5].

Karp's attack on frontier AI labs didn't appear out of nowhere. On July 1, a month before earnings, he told CNBC that AI models had been 'completely, irresponsibly oversold' and criticized labs' token-based pricing as extracting enterprise value for little measurable return ^[6]. By the Q2 shareholder letter and earnings call, the critique had sharpened into ideological language: labs building large language models, Karp wrote, intend to 'capture the means of production' of their enterprise partners, a dynamic he called having 'Marxist overtones and undertones' ^[1]. On the call itself he went further, telling listeners that enterprises are effectively paying frontier labs 'for the right...to migrate your IP, your know-how, your expertise to their model,' before adding the mocking line that such labs believe enterprises 'deserve to be colonized' ^[2]. The through-line across both interviews is the same: enterprises hand over proprietary data and operational expertise to train someone else's model, without owning any of the resulting intelligence - the inverse, in Karp's telling, of Palantir's model-agnostic approach where clients retain control of their own data and logic.

The reaction split sharply by audience. Financial commentary treated the numbers as unambiguous vindication of Palantir's enterprise-AI thesis over OpenAI and Anthropic. But the 'Marxist' framing itself drew open hostility - critics argued Karp inverted the term he was weaponizing, contending it's AI labs privatizing a public good and selling it back to customers that more closely resembles the dynamic Marx actually described, not the reverse. That skepticism sat alongside a second, older tension: investors who are bullish on the stock but openly uncomfortable with Palantir's surveillance and government-contract business, holding shares on the numbers while admitting distaste for what the company does with them. The combination - celebrating the growth, distrusting the rhetoric, and staying wary of the underlying business - meant the earnings cycle amplified Karp's voice without necessarily converting skeptics.

The timing isn't incidental. As OpenAI and Anthropic push further into enterprise-facing products, Karp's rhetoric works to draw a bright line between two competing models for enterprise AI: a model-agnostic, customer-owned-IP approach versus one where proprietary data trains vendor-controlled foundation models ^[2]. Coverage around the dispute also points to a broader shift in enterprise behavior - companies auditing AI usage and cutting spend that isn't tied to measurable ROI, a trend Karp's commentary both reflects and reinforces ^[7]. That some in venture capital, including Marc Andreessen, amplified Karp's July critique suggests the 'enterprises are being IP-farmed by frontier labs' framing has audience beyond Palantir's own shareholder base ^[7], even as OpenAI and Anthropic themselves have not directly responded in the research reviewed here.