Intel, Cray land $200M US Department of Energy contract for building two supercomputers based on Xeon Phi chips
Ryan Smith / AnandTech :
Context & Ripple Effects
This 2015 award is the seed of the Intel–Cray line of Department of Energy machines: the same pairing returns four years later when Intel claims it will deliver America's first exascale supercomputer by 2021 on its Xe architecture, and DOE's broader hedging strategy shows up in a $258M R&D award shared among six vendors including AMD, IBM and Nvidia. The contract also lands Xeon Phi at a contested moment — Intel had just pitched next-gen Xeon Phi for deep learning while Nvidia accused it of marketing on old benchmarks.
For Intel, the deal is a reference-customer play: DOE gives Xeon Phi a marquee deployment that counters Nvidia's dominance in accelerator-based systems, while Cray converts its systems-integration franchise into recurring federal business — a franchise valuable enough that HPE later buys the company outright.
First-order effects
- Intel gets two flagship DOE deployments built on Xeon Phi, giving the chip a national-lab reference against Nvidia just as their deep-learning benchmark spat is playing out publicly.
- Cray locks in design and integration revenue from a $200M federal program, reinforcing its position as the go-to builder of DOE systems ahead of the larger contracts that follow.
Second-order effects
- Nvidia faces a subsidized competitor in the exact market segment — accelerator-heavy scientific computing — where it had been setting the terms, pushing both vendors toward the DOE R&D money and benchmark credibility contests.
- DOE learns from single-vendor exposure: the later $600M Cray machine and HPE's decision to put AMD silicon into that system show the agency and its integrators deliberately spreading accelerator bets beyond Intel.
Third-order effects
- If the pattern holds, DOE procurement functions as the de facto funding mechanism for US accelerator roadmaps — vendors use federal contracts to prove architectures, then carry those references into commercial AI/data-center sales like Intel's $1B Xeon AI business.
- The structural endpoint is multi-vendor supercomputing as standard policy: every subsequent major DOE award splits silicon suppliers (Intel, AMD, Nvidia), making 'who builds' (Cray/HPE) separate from 'whose chips' — a hedge against any one vendor's roadmap slipping.
The trend: Federal supercomputing contracts have become the proving ground where US chipmakers fund and validate accelerator architectures, with DOE deliberately rotating silicon vendors to hedge each generation's execution risk.