The Mitre Corporation is to deploy a GPU supercomputer from Nvidia for government use.

Nvidia this week announced the non-profit firm is implementing a new DGX SuperPod system for use by US federal agencies.

DGX H100 SuperPod
– Nvidia

The deployment – described as Mitre’s flagship supercomputer – will support the company’s Federal AI Sandbox, a platform to improve experimentation with AI-enabled applications across federal government agencies.

Nividia said the SuperPod will offer a performance capacity of 1 exaflops of AI performance. The Federal AI Sandbox is expected to be operational by year’s end.

“Mitre’s purchase of a DGX SuperPod will help turbocharge the US federal government’s development of its AI initiatives,” said Anthony Robbins, vice president of public sector at Nvidia. “AI has enormous potential to improve government services for citizens and solve big challenges, like transportation and cybersecurity.”

Mitre was formed in 1958 as a military think tank spun out from the radar and computer research department at the MIT Lincoln Laboratory. Today it is a major supplier of services to the US government and its intelligence agencies.

The company said the sandbox will be applied to work for federal agencies in areas including national security, healthcare, transportation, and climate. It can be used to train AI applications including large language models. Agencies will be able to access the sandbox through existing contracts with any of the six federally funded research and development centers Mitre operates.

"The recent executive order on AI encourages federal agencies to reduce barriers for AI adoptions, but agencies often lack the computing environment necessary for experimentation and prototyping," says Charles Clancy, Mitre SVP and CTO. "Our new Federal AI Sandbox will help level the playing field, making the high-quality compute power needed to train and test custom AI solutions available to any agency."

Mitre and Nvidia didn’t say where the SuperPod would be hosted, but the Washington Post reports it will be hosted in Ashburn, Virginia. Mitre operates a data center in Bedford, Massachusetts, and has a large presence in McLean, Virginia.

The WaPo said the system will consist of 256 Nvidia GPUs, at a cost of $20 million.