Nvidia and AMD focus on artificial intelligence sparks computing supply concerns

Sandia National Laboratories, which operates supercomputers at Kirtland Air Force Base in New Mexico, relies on chips from Nvidia and AMD for complex simulations like nuclear weapons modeling. However, the companies’ shift toward AI-focused chips has created supply shortages, raising concerns about meeting the lab’s demands for double-precision floating point computations, critical for scientific accuracy. The lab’s high-performance computing team, led by Steve Monk, has expressed stress over securing suitable chips, prompting exploration of alternatives. NextSilicon, an Israeli startup, has developed chips using a data flow architecture distinct from Nvidia’s GPUs and AMD’s CPUs. These chips passed initial supercomputing tests in collaboration with Penguin Solutions and could undergo further evaluation for nuclear security applications later this year. Nvidia and AMD have historically dominated supercomputing contracts, but their AI-driven priorities have reduced double-precision performance in newer products like Nvidia’s Rubin chips. Analysts note this shift may disadvantage scientific computing, though Nvidia claims its chips remain balanced for both AI and real-world applications. Sandia’s decision to test NextSilicon’s chips reflects broader market changes, where AI demand is opening opportunities for smaller firms. The startup’s chips, designed for efficiency and adaptability, could address Sandia’s needs if they meet rigorous testing standards for nuclear security simulations. Officials at Sandia emphasize the urgency of securing reliable computing power, highlighting the unintended consequences of the AI chip race. The lab’s collaboration with NextSilicon marks a potential turning point in diversifying its supply chain and ensuring mission-critical capabilities remain intact.