National Labs Turn to Chip Startups as AI Boom Strains Supply
Sandia National Laboratories, which operates supercomputers for nuclear weapons research on Kirtland Air Force Base in New Mexico, is testing chips from NextSilicon, an Israeli startup. The move reflects a broader shift: as major chipmakers like Nvidia and AMD prioritize artificial intelligence, government labs are struggling to secure processors for scientific computing work that requires different technical capabilities.
For over a decade, Sandia relied on mainstream semiconductor firms for the computing power needed to simulate how hypersonic nuclear weapons move through the atmosphere or model nuclear detonations. That supply line is tightening.
"The pressure we're feeling right now is on the computing front and also from the supply chain," said Steve Monk, manager of Sandia's high-performance computing team. "Looking to the future, it's a bit stressful in terms of our ability to deliver to the mission."
The Double-Precision Problem
The core issue centers on double-precision floating point computation-the ability to calculate both very large and very small numbers without rounding errors. Physics simulations and nuclear security research depend on this capability. AI applications do not.
Nvidia's forthcoming Rubin chips show declining double-precision performance by some measures, according to Ian Cutress, chief analyst at More Than Moore, a chip consulting firm. AMD is releasing a scientific computing variant of its processors, but the market shift is clear: AI work now drives chip design priorities at the industry's largest players.
Daniel Ernst, senior director of supercomputing products at Nvidia, said the company remains committed to scientific computing and aims to create balanced chips. The reality, however, has prompted Sandia to evaluate alternatives.
NextSilicon Passes Key Test
On Monday, Sandia announced that NextSilicon's chips passed a technical milestone in general supercomputing tests. The chips use a data flow architecture fundamentally different from the graphics processing units (GPUs) and central processing units (CPUs) that dominate the market.
NextSilicon's approach saves electricity by reducing the amount of data shuffled between the processor and memory. The chips also reprogram themselves on the fly to run more efficiently. If testing continues as planned, the systems could advance to evaluation with actual nuclear security workloads this fall.
Building a Backup Supply Chain
James Laros, a senior scientist at Sandia overseeing the program to test new computing architectures, said the work with smaller players ensures the lab can procure chips even if major firms shift focus.
"We have to keep available options to complete our mission, because the mission is not optional," Laros said.
Sandia's influence on the chip industry extends beyond procurement. The lab's advocacy for liquid cooling systems more than a decade ago-once considered exotic-helped make the technology standard across the industry. That history suggests NextSilicon's success could open doors for other newcomers in scientific computing.
The shift also reflects how market concentration in AI chips is creating unexpected opportunities for startups. When the largest players chase the same market, niche applications become available to smaller competitors willing to meet specialized needs.
Learn more about AI for Science & Research and how computational advances shape research workflows.
Your membership also unlocks:
