Palantir Technologies Inc. (NASDAQ:PLTR) is widely treated as an artificial-intelligence stock, but one defense AI CEO says its real strength is the “plumbing” around AI models, not the models themselves.

“They’re not really an AI company. They’re an incredible data architecture company,” Tyler Saltsman, CEO of EdgeRunner AI, told Benzinga.

Saltsman called Palantir “a great partner” and said EdgeRunner has begun testing its military-specific AI models with Palantir’s Maven system.

Palantir Provides the AI ‘Plumbing’

Saltsman said Palantir’s strength lies in the engineering and data infrastructure around AI rather than building foundation models itself.

“AIs are really only as good as the engineering around them,” he said.

He compared EdgeRunner to the “camshaft” and Palantir to the “engine block,” with EdgeRunner supplying specialized models that can use Palantir data.

“We’re an awesome add-on to Palantir,” Saltsman said.

EdgeRunner Wants Maven Data Off the Grid

EdgeRunner builds smaller models designed to run directly on military devices without an internet connection.

Saltsman said its models can pull Palantir Foundry data onto a device, organize it and give soldiers the most important information locally.

“We’re starting to test it now, and then yes, it’ll be the standard in the future,” he said.

Saltsman argued that Palantir “doesn’t operate without the net,” making an offline layer important when troops lose connectivity.

EdgeRunner Eyes a Bigger Future

Saltsman said there is “no reason” every U.S. warfighter should not eventually have EdgeRunner as part of their kit.

Saltsman sees EdgeRunner expanding beyond defense into regulated industries, vehicles and other on-device AI applications.

An IPO is “the dream,” he said, though any listing could still be roughly 10 years away. Saltsman pointed to Anduril as evidence a successful defense-tech company can remain private for years.

That broader ambition fits Saltsman’s bet on smaller, local models. He told Benzinga this week that the AI data-center buildout may be “overkill,” arguing on-device models could eventually reduce demand for centralized compute.

Maven Faces Scrutiny

Palantir Technologies’ Maven Smart System has faced scrutiny following the Feb. 28 strike on a school in Minab, Iran, that killed at least 123 children. Officials involved in an internal Pentagon review told Bloomberg that U.S. forces carried out the strike and that flawed intelligence, outdated satellite imagery and overreliance on AI contributed to it. Some personnel expected Maven to flag stale or inconsistent intelligence, but it did not.

Palantir said it was not responsible for the underlying intelligence and there was no evidence its software caused the strike.

Saltsman said the incident showed why military AI needs continuously updated data rather than relying on information that may no longer reflect conditions on the ground.

Prediction-market traders remain focused on the model race. Polymarket gives Anthropic a 74% chance of having the best AI model at year-end, compared with 13% for OpenAI and 11% for Google, on about $1.4 million in volume.

Saltsman’s argument suggests Palantir may not need to win the model race. If specialized models increasingly plug into Maven and Foundry, Palantir could remain central by providing the infrastructure around them.

Image: Shutterstock