Why Enterprises Are Unhappy With Frontier AI Labs: Palantir CEO Explains

The “Tokenmaxxing” Problem

Karp put a name to the frustration: tokenmaxxing.

The idea is that frontier labs are incentivized to burn through as many AI tokens as possible — treating high token consumption as a proxy for productivity and value. Enterprises, meanwhile, are watching their AI bills climb and wondering what they’re actually getting for it.

It’s a classic misalignment. Labs are optimizing for scale and model usage. Businesses are optimizing for outcomes. Those two things are not the same.

Accelerating costs are already raising alarm on Wall Street, and the concern is straightforward: as AI gets embedded deeper into enterprise workloads, the cost curve doesn’t flatten — it steepens.

“The Implementation Is Where the Value Is”

Karp was careful not to dismiss large language models entirely. His critique is more surgical.

“It is not that large language models aren’t crucial for the world,” he said. “It’s just the implementation is where the value is, certainly in the next seven years.”

That framing is essentially Palantir’s entire pitch — and it’s a credible one. Raw model capability matters less than how that capability gets deployed inside complex, regulated, high-stakes enterprise environments. Anyone who has tried to take a frontier model from demo to production knows the gap is real.

Karp also dropped a notable aside: most of Anthropic‘s publicly discussed projects are “running on Palantir.” Make of that what you will, but it suggests the infrastructure layer is where the quiet leverage lives.

The IPO Backdrop

Karp’s comments land at an interesting moment. Both Anthropic and OpenAI are moving toward public markets — Anthropic filed confidentially, and OpenAI followed a week later with its own confidential IPO filing.

Going public means scrutiny. Revenue quality, customer retention, and enterprise satisfaction will all come under the microscope. If Karp’s read is accurate — that enterprise frustration is widespread and private — that’s a meaningful headwind for labs trying to tell a clean growth story to public market investors.

He acknowledged Anthropic’s Dario Amodei as “a very, very important person” leading “the leading frontier model company,” even while noting they often disagree. Competitive respect, diplomatically delivered.

What This Means for Enterprise AI Buyers

If you’re evaluating AI tools for your organization right now, Karp’s comments are worth sitting with.

The frontier labs are extraordinary at building models. They are less focused on whether those models actually integrate cleanly into your workflows, comply with your governance requirements, or justify their cost at scale. That gap — between model capability and enterprise implementation — is where a lot of real-world AI projects quietly stall.

The smarter move isn’t to chase the most powerful model. It’s to ask which platform actually helps you deploy, monitor, and extract value from AI in your specific context.

The Bigger Picture

Karp also expressed frustration at the politicization of AI broadly, arguing the technology will drive the most consequential decisions in America’s near future. “This is a massive revolution,” he said, “and there are opportunities only America has, and there are dangers in this revolution.”

Whatever you think of Karp’s politics — and they are genuinely complicated — his read on enterprise AI sentiment is hard to dismiss. He talks to a lot of large organizations. They’re telling him something consistent.

The frontier labs built the engine. Now enterprises want someone to actually drive the car.