Booz Allen Moves Its Defence and Cyber Software From Anthropic Models to Nvidia's Nemotron
The government contractor's technology chief says the switch to open models has cut costs, but its developers still mostly write code with Anthropic's Claude Code.
By Nakoda News Desk
·2 min read
Prefer Nakoda News on Google
Booz Allen Hamilton switched the software it runs for cybersecurity and defence customers from mainly Anthropic's models to Nvidia's open Nemotron models several months ago, chief technology officer Bill Vass told The Information in an interview. Vass said the change had produced significant cost savings. The company did not give figures.
The switch gives Nvidia a high-profile enterprise customer for its own models at a difficult time for the chipmaker, whose board is investigating chief executive Jensen Huang. It is also an example of a large customer moving from a frontier lab to cheaper open models once AI costs became significant. Booz Allen, a government contractor with a market value of about $10 billion, spends tens of millions of dollars a year on AI, according to a person familiar with the company cited by The Information.
"What we've been doing is the opposite of tokenmaxxing," Vass said, using the industry term for using as much AI as possible.
The change covers the AI that runs Booz Allen's products, not the tools its engineers use to build them. Vass said the company's developers still mostly use Anthropic's Claude Code unless customers ask for something else. He said he would like developers to use Nemotron for coding as the models improve.
Nvidia has been selling Nemotron to government agencies and contractors partly through Palantir. In late June, Palantir launched a product to help US agencies and businesses run Nvidia's open-source models on its software, and Nvidia has signed a deal to use Palantir software to manage its own supply chain. Both companies present themselves as a way for customers to avoid depending on Anthropic and OpenAI. Palantir chief executive Alex Karp told The Information this summer that some government customers had already moved from closed frontier models to Nemotron.
Usage data shows Nemotron gaining in some places but still far behind the frontier labs. Kilo Code, an open-source coding platform, said Nemotron models handled nearly 17% of the roughly 50 trillion tokens processed through its software in October, up from 8% the previous month. On OpenRouter, which processes about 700 trillion tokens a month for a wider range of tasks, Nemotron accounted for about 4% of tokens over the past month, down from a weekly peak of 7% in August. So far in November, Anthropic has taken 31% of OpenRouter customers' spending, more than any other provider, while Nvidia's share was below 1%.
The cost pressure behind Booz Allen's decision is also showing up in hiring. Maria Colacurcio, chief executive of HR software company Syndio, said recruiters at large companies are getting requests from job candidates for higher or unlimited token budgets and access to the latest OpenAI and Anthropic models as part of their offers. She said these arrangements are not yet written into formal policy.
Vass did not say when Booz Allen might move its coding work to Nemotron. That depends on how quickly Nvidia's models catch up with Claude Code on software engineering.
Sources
- The Information · Booz Allen Switches to Nvidia's Nemotron
Produced with AI assistance from the sources above. How we work
By
Nakoda News Desk
Nakoda News covers AI and business. Every story links the sources it draws on; how we work is set out in our editorial standards.


