THE ESSENTIALS IN 30 SECONDS
- China's CPU-only LineShine stays No. 1 at 2.198 exaflop/s, ahead of El Capitan (1.809), and the top 10 are unchanged since June.
- Only 27 systems are new, the fewest in the list's history; organizers say new hardware is going mainly into large AI clusters.
- AMD says it powers 202 systems; Oak Ridge's Lux, an AI-for-science machine with Instinct MI355X GPUs, enters at No. 14 with 220.79 petaflop/s.
The 68th edition of the TOP500 ranking of supercomputers was published on November 16 at the SC26 conference in Chicago. The headline is stability at the top and an unusually quiet list overall, which the organizers link to where new computing hardware is now being installed.
A frozen top 10
LineShine, at the National Supercomputing Centre in Shenzhen, keeps the No. 1 spot it took in June, when it became the first China-based leader since Sunway TaihuLight in 2017. It reaches 2.198 exaflop/s on the HPL benchmark using only CPUs: custom 304-core LX2 processors linked by the LingQi interconnect, drawing about 42 megawatts. It also leads the HPCG benchmark at 22 petaflop/s. Behind it, El Capitan (1.809 exaflop/s), Frontier, Aurora and Germany's JUPITER Booster complete the five exascale systems, and every top-10 machine holds its June position. El Capitan still leads the mixed-precision HPL-MxP ranking at 16.7 exaflop/s, according to Network World.
Fewer new systems, more AI clusters
Only 27 systems are new to the list, the lowest count in the ranking's history. The list still grew: total HPL performance rose 4.4% to 19.57 exaflop/s and the entry threshold rose 7.3% to 2.857 petaflop/s, Network World reports, while 291 systems use accelerators. The organizers say much of the new hardware is going into large AI clusters, a move from capability systems towards what they call "AI factories." They also cite an Epoch AI estimate that five US hyperscalers owned 71% of global AI compute capacity at the end of 2025. Such private clusters rarely submit benchmark results, so the TOP500 increasingly reflects only part of the world's compute.
AMD and Lux
AMD says its chips are in 202 systems, over 40% of the list and about 14% more than a year earlier, and in half of the new entries. It powers four of the top 10. The most notable newcomer is Lux at Oak Ridge National Laboratory, which debuts at No. 14 with 220.79 petaflop/s. Built by HPE with AMD EPYC CPUs, Instinct MI355X GPUs and Pensando networking, it came out of a public-private partnership between AMD and the US Department of Energy. AMD presents it as the first dedicated AI system for the DOE's Genesis Mission, with early projects in fusion power plant design, a foundation-model effort for materials, and water forecasting in the Tennessee Valley. AMD has not published Lux's GPU count.
