DOE awards $258M to be shared among AMD, Cray, HPE, IBM, Intel, and Nvidia over next three years for R&D into exascale computers to better compete with China
Rachael King / Wall Street Journal :
Context & Ripple Effects
The DOE has been building toward this for years: in 2015 it handed Intel and Cray a $200M contract for two Xeon Phi-based supercomputers while separately outlining national-lab computing centers with IBM and Nvidia (the 2015 centers collaboration). The $258M R&D pool now spreads pre-exascale development across all six vendors at once, explicitly framed as competition with China.
What makes the award consequential is how the pipeline played out afterward: Cray won a $600M DOE system for nuclear-weapons simulation while being acquired by HPE, which then brought AMD's chips into that machine, and by 2025 the DOE was accelerating lab AI supercomputers with Nvidia, AMD, and Oracle sharing costs (the 2025 cost-sharing push). This 2017 grant is the seed round of that procurement arc.
First-order effects
- AMD, Cray, HPE, IBM, Intel, and Nvidia each get a slice of three years of federal R&D funding, letting them de-risk exascale architectures ahead of actual system procurements like the Intel-Cray build already underway.
Second-order effects
- The multi-vendor structure keeps DOE contracts contestable — which is exactly what let HPE's acquisition of Cray and its AMD partnership displace an Intel-centric lineup on the later $600M award.
Third-order effects
- If the pattern holds, the DOE functions as a standing anchor customer that shapes US high-performance computing architecture and consolidates the vendor field around fewer, larger systems integrators — with vendors increasingly asked to co-fund the machines they build.
The trend: Washington is treating frontier compute as strategic leverage against China, with the DOE evolving from grant-maker to anchor customer whose procurement choices pick winners in US supercomputing.