The TOP500 list of the world’s fastest supercomputers is published twice a year, and it usually shows a steady stream of new machines. The November 2026 edition, released at the SC26 conference, is different: only 27 new systems entered the list. According to The Next Platform, that is the lowest turnover across 68 editions since the list began in 1993.
The biggest newcomers
The largest new entry is Hangang at KISTI, the Korea Institute of Science and Technology Information. It is an HPE Cray EX254n system using NVIDIA Grace-Hopper chips and Slingshot 400 networking, with a theoretical peak of 588.3 petaflops.
The second-largest newcomer is Lux at Oak Ridge National Laboratory in the United States, built with AMD MI355X GPUs and EPYC 9575F processors and rated at 323.9 petaflops.
A cloud system with many cores
One unusual new entry is a China Telecom cloud system. It accounts for 93.3 percent of all the new CPU cores added to the list, but only 2.9 percent of the new peak performance, showing how much more work accelerators do per core.
Why have installations slowed?
The Next Platform offers several possible reasons. Buyers may be waiting for the next generation of chips, networking is still at 800 Gb/s, and some organizations may prefer renting cloud capacity. The author leans toward roadmap timing. The TOP500 editors stress that the slowdown does not mean performance has stopped growing.
Accelerators dominate
Accelerated systems make up 290 of the 500 machines, or 58 percent, and deliver about 80 percent of the list’s total peak performance. Measured by double-precision performance, NVIDIA accounts for 52.9 percent and AMD for 37.7 percent.
Looking ahead
New chips such as NVIDIA’s Rubin HPC, announced at the same conference for 2028, could restart the cycle of new installations. For now, the November 2026 list is a snapshot of a pause, not a decline.
Sources:
