As artificial intelligence (AI) reshapes national missions and data strategies, public and private leaders are building a new class of infrastructure – AI factories – to power large-scale model development and secure, sovereign deployments. At the recent NVIDIA GTC conference in Washington, D.C., technologists described AI factories as the next foundation for both commercial innovation and government modernization.
Federal agencies are already exploring how purpose-built AI infrastructure can accelerate modernization while meeting requirements for sovereignty, sustainability, and compliance, in line with The White House’s AI Action Plan. That vision echoes what industry leaders say is happening on the ground: a convergence of data center design, policy, and national competitiveness.
“We are in a state today where we’re putting more demand on institutional frameworks than we’ve ever put before,” said Ken Patchett, vice president of data center infrastructure at Lambda, which builds gigawatt-scale AI factories.
Patchett described AI factories as the critical backbone of that infrastructure – purpose-built, high-density environments designed for deterministic workloads, not elastic, multi-tenant clouds.
The AI factory approach is being driven by three forces: a surge in enterprise AI adoption, a rise in open-source model development, and growing sovereign demand, experts said.
“There is real uptick in enterprise use cases around … customer support, voice models, e-commerce … and huge sovereign demand. Countries are starting to build their own AI factories for their own use cases, for their own citizens,” said Mahadev Konar, senior vice president of infrastructure engineering at Together AI. The company’s AI cloud platform helps developers and researchers train, fine-tune, and deploy generative AI models.
Open infrastructure is essential to maintaining control of data and models, said Ioannis Antonoglou, co-founder and chief technology officer of ReflectionAI, which builds open foundation models for software development. “If we don’t do it, and if there isn’t a really frontier open-weight model coming out of the U.S., then it’s either you export this technology or you import it,” he said. “For both countries and enterprises, it’s important that there is an alternative that ensures we shape the new protocol and infrastructure of AI.”
Jeff Jones, head of revenue, Americas at generative AI startup poolside, noted that his company’s products are built “from day one” to support U.S. government and defense missions, with security and auditability engineered into the deployment model.
“It’s been very critical to us to be able to adapt to where [software] needs to be – whether that’s within data centers, on bare metal, or in the field. You have to support the innovation those missions want … but also the security and compliance requirements as well.”
Sustainability and local engagement are pillars of future AI infrastructure, experts said. As data center development expands to meet the demands of enterprise AI, success depends on public-private collaboration at every level, “from the schools all the way through the state and federal level,” Patchett said.
Building AI factories, he said, industry to help communities understand that these facilities are “the new factories” that create long-term jobs, and to ensure that “every electron we consume should be done for a reason.”