Who Really Makes Money in AI? Palantir’s CFO has an Answer

In an interview at Palantir’s AIPCon, CFO David Glazer declared that “LLMs are commodity cognition”. His point was that the raw output of an LLM is getting cheaper and better (“ELO scores [are] up, tokens are cheaper”) so the real challenge is how companies derive value from that output. “We are the product, we have the data, we have the customer relationship and we can vend in whatever intelligence sources we need in order to accomplish the task,” he said.

This framing happens to highlight a crucial divide in AI. Foundation models are improving fast, but as their costs fall and supply expands, they look increasingly like interchangeable infrastructure. The companies that turn those models into workflow tools and integrate them with data are the ones showing durable margins and higher valuations.

Commoditization in Motion

There is evidence that LLMs are already sliding toward commodity status. Analysts at Summit Partners describe the foundation model market as “fiercely competitive, as companies push for better performance at lower costs, often with shrinking margins”. IoT Analytics reaches a similar conclusion: “the availability of more and cheaper models will ultimately lower the cost of including GenAI features in applications”.

That dynamic means LLMs look more like cloud servers or GPUs; vital, but fungible. In financial terms, investors are already pricing this in. In February this year, leading model developers such as OpenAI and Anthropic were valued at lower revenue multiples than they were a year ago.