US Department of Energy says AMD and Cray will build the world's fastest supercomputer, dubbed Frontier, by 2021, with 1.5 exaflops of processing power
Dubbed Frontier, the supercomputer expected to go online in 2021 with 1.5 exaflops of processing power
Context & Ripple Effects
Two months after Intel announced it would build America's first exascale machine using its new Xe GPUs alongside Cray (Intel's exascale claim), the Department of Energy has handed the same integrator a second exascale contract built on AMD silicon instead. Frontier, targeted for 2021 at 1.5 exaflops, makes AMD rather than Nvidia or Intel the compute engine of the US flagships program.
That bet paid off on the record books: Frontier topped the Top500 as the world's fastest system (overtaking Japan's Fugaku) and held the crown into late 2023 above Intel-powered Aurora, with 121 of the world's fastest machines running AMD silicon by mid-2023 (121 systems on AMD silicon) — momentum that culminated in the 2025 $1B Lux and Discovery partnership with the same customer.
First-order effects
- Intel loses its position as sole declared US exascale vendor: its Xe-architecture bid now competes inside the DOE portfolio against an AMD-Cray machine with the same deadline, and Cray emerges as the neutral integrator serving both chip camps.
- AMD converts a research-lab design win into flagship validation, giving its server silicon a reference deployment no competitor can match on paper.
Second-order effects
- DOE effectively runs a two-vendor exascale race, forcing Intel to answer through its own Argonne-class system — the Aurora machine that later appears in the rankings behind Frontier — rather than ceding the segment.
- Cray's vendor-agnostic integration role lets it win either way, while AMD's HPC credibility feeds the broader datacenter business reflected in its later Data Center revenue growth.
Third-order effects
- Government flagship contracts are functioning as credibility engines: the Frontier win seeded the broader Top500 penetration and, if the pattern holds, positions AMD as the default partner when Washington scales up again, as it did with Lux and Discovery.
- Exascale procurement shifts from one-off national bets toward sustained multi-year compute partnerships between the state and a small set of chip vendors, making Top500 standing a proxy for national AI capacity.
The trend: US exascale computing is moving from single-vendor flagship bets toward durable multi-system partnerships anchored on AMD, with DOE procurement setting the pace for the global Top500.