NASA’s Ames Computing Expansion Connects Scientific Research and AI

NASA announced September hardware additions to two computing systems at its Ames Research Center in California, extending an established scientific-computing campus that also supports artificial intelligence. The operational notices described two new racks for the CPU-based Athena supercomputer and 188 additional Grace Hopper nodes for the GPU-based Cabeus system. An October 13 NASA notice confirms that the 188 new Cabeus nodes are now available to all users. Completion of Athena’s September rack integration remains unverified in the reviewed documents.

The distinction matters because Ames combines different kinds of computing. Athena provides high-performance CPU resources for research, while Cabeus is explicitly designed for applications that benefit from GPUs, including machine learning. Calling the entire expansion a new AI data center would obscure both the existing campus and the separate hardware deployments.

An operating system, followed by planned additions

NASA announced Athena’s availability on January 27, 2026, describing a January rollout to existing users after beta testing. Its technical page also records an initial release at the end of December 2025. Those accounts describe different stages of access; the January announcement establishes the agency’s public account of the rollout without making September’s later additions operational by implication.

DatacenterDynamics reported on January 29 that Athena had come online and replaced Pleiades. Its article attributed the architecture figures to NASA. That reporting provides independent editorial coverage of the launch, but it does not independently verify the more recent upgrade work.

A September 4 NASA notice scheduled partial downtime on September 22–24 to begin connecting two additional racks, followed by a full outage on September 29 to complete integration. A separate September 14 notice scheduled September 17 downtime to integrate 188 new Grace Hopper nodes into Cabeus. Those September notices were work plans. NASA subsequently confirmed the Cabeus additions were available to all users on October 13; Athena’s elapsed work dates alone do not establish installation acceptance.

Where the suppliers fit

NASA’s published Athena configuration identifies Hewlett Packard Enterprise as manufacturer and AMD EPYC 9745 processors as the computing hardware. The table lists 1,024 nodes and 262,144 CPU cores. It remains a published configuration, rather than a fresh acceptance report covering September’s new racks.

For Cabeus, NASA identifies an HPE-manufactured A100 section and a Supermicro-manufactured GH200 section using NVIDIA technology. NASA’s October 13 notice and current technical page report 538 GH200 nodes, including the 188 nodes added in September. This is an agency-reported available configuration, not a measure of utilization.

These records establish specific equipment roles for HPE, AMD, NVIDIA and Supermicro. They do not disclose the upgrade purchase prices, supplier margins, payment dates or revenue attributable to the work. Nor do they establish that the manufacturer of an earlier configuration received a separate contract for every later addition.

Computing performance is not campus capacity

TOP500’s June 2026 record lists Athena at 15.98 petaflops on the Linpack benchmark, compared with a theoretical peak of 20.13 petaflops. Those are different performance measures, not electricity capacity or a guarantee of the speed of AI training. The listing also predates the September expansion notices.

The physical setting is established infrastructure. NASA opened the Modular Supercomputing Facility in August 2019, initially housing Aitken. Its opening account described outdoor air, fans and a circulating water system adapted to the Bay Area climate. Infrastructure designed for up to 16 modules was an expansion envelope, not evidence that all 16 were built.

NASA’s January Athena announcement reported lower supercomputing utility costs and described a hybrid approach combining agency supercomputers with other resources, including commercial cloud platforms. The reviewed evidence does not quantify current campus savings, establish a project-specific power-service agreement or identify a new cloud-provider contract.

What the next records need to establish

The next material evidence would be confirmation of Athena’s September additions and the further 266 Cabeus nodes NASA plans to integrate later in October, followed by procurement records identifying award values and scope. Usage and workload results would show how researchers are applying the resources; available hardware alone does not measure utilization.

For investors, Ames documents a government research customer using several publicly traded suppliers. The financial significance of this particular expansion remains unquantified. Its scientific and AI purposes are supported by NASA’s disclosures, while Athena’s expansion acceptance, further Cabeus additions and supplier economics require additional records.