Palantir CEO Alex Karp says some U.S. government customers have begun switching from proprietary AI models to NVIDIA's open-source Nemotron models. The approach keeps sensitive work inside a trusted application layer rather than relying directly on OpenAI or Anthropic.
June 29, 2026 · Palantir × NVIDIA
U.S. Agencies Swap Proprietary AI for Open-Weight Models They Can Control
Palantir's CEO says some government customers have moved from OpenAI and Anthropic to NVIDIA's open-weight Nemotron models — keeping sensitive work inside sovereign, air-gapped environments they own end to end.
The driver: stop paying for "tokens that create no value" while your data walks out the door.
Data sovereignty
Keep the weights on hand and fine-tune with your own data
Air-gapped
Runs in classified environments isolated from external networks
Data flywheel
Continuously improve the model in-house over time
A sovereign engine for U.S. agencies
NVIDIA
GPU compute + Nemotron open-weight models
→
Palantir
AIP · Ontology · Foundry · Apollo — on-prem, CMMC L2 / ISO
→
Sovereign engine
Air-gapped, classified, agency-controlled
Nemotron 3 family — model sizes
Approx. parameter count, drawn to scale (each block ≈ 100B).
~30B
Nano
Small & efficient · sub-agents
~110B
Super
High-accuracy · multi-agent
~525B
Ultra
Large-scale · complex tasks
99.2%
AIME 2025 with tool use
1M
tokens of context length
Why agencies favor it
Data stays inside controlled, isolated environments
Weights owned and continuously improved in-house
Framed as a foundation for national security & U.S. tech leadership
Open questions
Critics warn of creeping surveillance in defense work
Pricing and specific availability not yet detailed
Announced late June–July 2026 — production feedback still limited
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…