When an AI company makes a billion dollars in a single quarter, people listen to whatever its CEO says.

Palantir reported Q2 2026 earnings: quarterly profit topped $1 billion for the first time, revenue grew over 30% year-over-year, and both government and commercial segments beat expectations. The stock jumped over 8% after hours on the release. But the earnings figures themselves aren't the biggest news — the biggest news is CEO Alex Karp's tirade on the earnings call.

"Frontier Labs Are Too Untrustworthy for Enterprises"

Karp didn't pull punches. Frontier AI labs (referring to companies like OpenAI and Anthropic) are "too untrustworthy" for enterprise clients, he said — their business models and technical roadmaps are disconnected from what enterprises actually need. He even used a loaded term: "There's a Marxist tendency in this industry — the belief that if the technology is good enough, all problems automatically solve themselves. But enterprise clients don't care how good your technology is; they care whether it solves problems, whether it breaks, and who's responsible when it breaks."

Words carry weight coming from Karp. Palantir isn't some AI upstart — it's been doing enterprise software for 20 years, serving the Pentagon, CIA, Wall Street banks, customers with the highest reliability requirements on the planet. Palantir's AIP (AI Platform), launched in 2023, has become one of the most successful enterprise AI implementations — not because it uses the strongest model, but because it solves what enterprises care about most: security, control, auditability, not breaking things.

Karp's core argument: frontier labs have gone too far in chasing "ever-stronger models," but what enterprise clients want isn't the strongest model — it's the most reliable solution. A model that's 99% accurate but occasionally hallucinates might be fine for consumer use cases, but in intelligence analysis, financial risk control, or medical diagnosis, that 1% error is fatal.

Two Camps of AI Commercialization

Karp's comments lay bare a growing divergence in the AI industry: those already making money and those still burning it are now different species.

One camp is companies like Palantir, Microsoft, NVIDIA — they're making real money from AI, thinking about customer needs, delivery quality, compliance risk, ROI. Their launches don't just talk parameters and benchmarks; they talk case studies, ROI, how they help customers save or make money.

The other camp is most frontier model companies — they're still burning investor cash, pursuing the limits of model capability, thinking about AGI, alignment, capability frontiers. Their launches are full of words like "revolutionary," "disruptive," "new era," but they can rarely articulate exactly how much money an enterprise customer will save or make using their product.

Neither camp is right or wrong; they're at different stages. But Karp's reminder has value: if you're an enterprise, don't get fooled by the "strongest model" narrative — figure out what problem you're solving first, then pick the right tool. If you're an AI company, don't get obsessed with benchmark leaderboards either — the only capability that matters is the one customers will pay for.

Of course, Karp has his own commercial axe to grind — Palantir doesn't train foundation models; it's a model "porter" and "integrator," wrapping models from various vendors into enterprise-ready solutions. Trashing foundation model companies serves Palantir's positioning. But setting aside self-interest, his core point holds: the last mile of AI to B is never a model capability problem — it's a delivery and trust problem.

See you tomorrow.