AI

Los Alamos–OpenAI Partnership Sees 20x Growth in AI Usage

Three national security labs now process 3 billion tokens weekly through secure supercomputer access to experimental models.

Omega Editorial · November 21, 2026 · 3 min read

Key takeaways

  • Weekly AI token throughput at Los Alamos grew 20-fold to 3 billion tokens between February and August 2026, with user accounts tripling across three national labs.
  • OpenAI deployed three experimental models—5.4 OpenAI, 5.4 OpenAI mini, and o3 OpenAI—to run entirely on Los Alamos' secure Venado supercomputer infrastructure.
  • Researchers are applying the models to certification workflows, simulation code translation, and digital-twin development for classified national security work.
  • Weekly successful AI requests jumped from 1,200 to more than 41,200 as adoption accelerated across Los Alamos, Lawrence Livermore, and Sandia laboratories.
  • Year two priorities include improving file handling, API reliability, and coding workflow support as labs prepare for next-generation computing platforms.

One year into a groundbreaking partnership, Los Alamos National Laboratory reports explosive growth in artificial intelligence usage across three national security facilities, with weekly token throughput jumping from 150 million to more than 3 billion between February and August 2026.

The collaboration brings OpenAI's experimental models into a secure computing environment accessible to researchers at Los Alamos, Lawrence Livermore, and Sandia national laboratories—collectively known as the tri-laboratories. The service, called Venado-GPT, runs entirely on the Venado supercomputer, which entered service in 2024 specifically designed to handle AI workflows.

Deployment details

OpenAI deployed three experimental models to the secure environment: 5.4 OpenAI, 5.4 OpenAI mini, and o3 OpenAI. The models were named in homage to J. Robert Oppenheimer, Los Alamos' first director. All three run on Venado's local graphics processing units connected through HPE Slingshot interconnect technology.

User adoption accelerated rapidly. The number of tri-laboratory accounts tripled during the measurement period, while weekly successful requests surged from approximately 1,200 to more than 41,200.

"We've seen tremendous growth in the tri-laboratories' use of Venado-GPT," Los Alamos Director Thom Mason said in the announcement. "Our partnership with OpenAI continues to enable new kinds of research as we apply the models to solving some of our nation's most pressing scientific and national security problems."

Mission applications

Researchers are applying the AI models to certification workflows, engineering parameter exploration, and surrogate model development. Los Alamos teams have paired Venado-GPT with URSA—an open-source agentic software package developed at the lab—to advance classified work.

"We've been using Venado-GPT models together with URSA teams of agents to support several key certification workflows," said Nathan DeBardeleben, senior high-performance computing architect at Los Alamos. Applications include document intelligence for certification tasks and reasoning about complex design choices.

At Lawrence Livermore, researchers are using the 5.4 OpenAI model to automate translation of simulation models between code bases. Digital-twin teams have deployed the technology for automated workflows and code development, according to Brian Giera, the lab's associate program director for AI and manufacturing.

Some applications emerged unexpectedly. Los Alamos researchers discovered they could use the service to identify kernel and driver changes during network maintenance operations.

Why it matters

This partnership represents a significant test case for deploying frontier AI models in classified government research environments. The 20-fold usage increase and rapid user adoption suggest the technology is solving real problems for national security researchers rather than serving as experimental infrastructure. As the U.S. competes globally in AI development, demonstrating secure deployment pathways for advanced models in sensitive government facilities could influence how other agencies approach AI adoption.

Roadmap

For year two, Los Alamos and OpenAI plan to enhance file attachment handling, strengthen API reliability, and improve support for complex coding workflows. The labs are also preparing for upcoming Mission and Vision computing platforms.

"Our work with Los Alamos applies frontier AI models in high-performance computing environments to support complex scientific and national research challenges," said Katrina Mulligan, head of national security partnerships at OpenAI. "We're proud to work alongside the national labs to accelerate scientific research, bolster national security readiness and drive public sector innovation."

These details were first reported by the Los Alamos Reporter.

  • #openai
  • #national laboratories
  • #high-performance computing
  • #government ai
  • #classified research
  • #supercomputing

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in AI