The U.S. Department of Energy is collaborating with Nvidia and Oracle to construct seven new AI supercomputers aimed at enhancing scientific research and fostering agentic AI for discovery. This initiative will feature two significant systems at the Argonne National Laboratory, creating the largest AI supercomputing infrastructure the DOE has ever established.
During the annual GPU Technology Conference, Nvidia CEO Jensen Huang revealed plans for the supercomputers, including Solstice, a system powered by 100,000 Blackwell GPUs. This will be the largest AI supercomputer for the DOE. When combined with another system at Argonne, Equinox, which consists of 10,000 Blackwell GPUs, the two interconnected supercomputers will deliver an impressive total of 2,200 exaFLOPs of AI computational capability.
Huang stated that the partnership aims to create an “engine for discovery” that will provide researchers with access to state-of-the-art AI infrastructure to propel advancements across various fields, including healthcare and materials science. He emphasized that the systems would not only enhance computational power for scientific research but also contribute to the development of “agentic scientists,” a concept that aims to improve research and development productivity through innovative AI applications.
While Equinox is anticipated to become operational next year, there is no specified timeline for the introduction of the Solstice supercomputer. In addition to the major announcements regarding Argonne, Nvidia also declared a new collaboration with Palantir to integrate accelerated computing technologies into Palantir”s AI platform, enhancing data analysis capabilities and introducing customizable AI agents.
Argonne plans to deploy three additional Nvidia-based systems named Tara, Minerva, and Janus, which will be accessible to researchers from other institutions, thereby broadening the reach of AI-driven supercomputing resources.
Meanwhile, Los Alamos National Laboratory in New Mexico is set to receive two new AI systems based on Nvidia“s Vera Rubin technology, despite having launched a significant AI supercomputer, Venado, just last year. The two upcoming systems, Vision and Mission, are scheduled for deployment in 2027. Vision is intended for unclassified research in national security, materials science, energy modeling, and biomedical fields, while Mission will focus on classified workloads, replacing the recently launched Crossroads supercomputer.
LANL director Thom Mason highlighted the importance of these new systems, stating that they represent a substantial investment in both national security science and fundamental scientific research capabilities. He noted that the Mission and Vision systems are specifically designed for supercomputing in the era of AI.
As the landscape of AI in scientific research continues to evolve, the introduction of these supercomputers may lead to significant advancements in various scientific domains. However, questions remain regarding the effectiveness of AI agents in achieving accurate results within research contexts.
