Booz Allen Hamilton CEO on AI in government: Space Llama, $1.5T defense ambitions, and the a16z partnership
Jan 13, 2026 with Horacio Rozanski
Key Points
- Booz Allen Hamilton positions itself as the largest AI provider to the federal government and stands to benefit from the Trump administration's proposed $1.5 trillion defense budget.
- The firm adapted Meta's Llama model to run on the International Space Station, demonstrating its core capability: making commercial AI functional in edge environments with no connectivity or stable power.
- Booz Allen formalized a partnership with Andreessen Horowitz to systematically identify and integrate technologies from a16z's 1,100-company portfolio into government missions, starting with Mistral on NATO defense problems.
Summary
Horacio Rozanski, president and CEO of Booz Allen Hamilton, has spent 35 years at the firm — joining as a summer intern and never leaving. That arc shapes how he talks about the business: less a consulting firm, more a technology integrator whose job is taking commercial innovation and making it work in environments where the commercial world never designed it to work.
Booz Allen is the largest AI provider to the federal government, with roughly three-quarters of its business tied to national security. The civilian quarter — VA, HHS, Treasury, IRS — is where budget pressure is most acute under the current administration. On the defense side, the direction is the opposite: the Trump administration is pushing toward a $1.5 trillion defense budget, and Rozanski frames Booz Allen as a natural beneficiary of that spend, provided Congress approves it.
Space Llama
The clearest illustration of what Booz Allen actually does is Space Llama. The firm took Meta's Llama model and adapted it to run on the International Space Station — the first large language model deployed there. Astronauts can now query equipment manuals in natural language rather than working through paper documentation. The adaptation challenge is the point: commercial AI assumes a data center with stable power, fiber connectivity, and 24/7 uptime. Booz Allen's proprietary layer makes these models functional at the edge — in deserts, underwater, in space, with no connectivity and constrained power.
The a16z partnership
About a year ago, Rozanski began conversations with Ben Horowitz and Marc Andreessen about formalizing what had been ad hoc collaboration. Booz Allen had already been working with a16z portfolio companies — Databricks on Department of Defense data problems, Shield AI's Hivemind autonomy platform as a foundation for Booz Allen's cybersecurity stack — without engaging a16z directly. The new partnership, announced Monday, creates a structured discovery and vetting process across a16z's portfolio of more than 1,100 companies.
The logic is that most of those companies are not building for government missions but are building technology that is applicable to them. Rozanski's team works problem-back: identify a mission gap, then find which combination of startups and internal Booz Allen tech can close it. Company size is secondary. Booz Allen runs its own corporate venture fund with positions in quantum firm Seek, low-earth-orbit satellite company Elido, ground autonomy startup Scout AI, and 3D-printed drone maker Firestorm.
One active example is Mistral, an a16z portfolio company, where Booz Allen is working on NATO and European allied defense problems — a set of requirements that became urgent as European governments sharply increase defense spending.
On hiring, Booz Allen is recruiting AI researchers and, notably, astrophysicists for its quantum team. The model is not to compete with OpenAI or Anthropic for foundation model talent but to build the integration and adaptation layer that sits on top of frontier models and makes them mission-ready.
Rozanski has known Jensen Huang since Nvidia was worth "a couple billion dollars" and the two companies have worked together since 2017. The a16z partnership follows the same logic: build the relationships at the platform level before the obvious commercial need makes them expensive.