NVIDIA says its technology powers more than 400 of the world’s top supercomputers, dominating new and energy-efficient deployments

NVIDIA used the release of the newest TOP500 and Green500 rankings at ISC High Performance in Hamburg to underline its grip on high-performance computing built for artificial intelligence and scientific workloads. In a company blog post, NVIDIA said its technology now powers more than 400 of the world’s 500 fastest supercomputers, reaching 81% of the TOP500 list.

The company also pointed to momentum among systems newly added to the ranking. Nearly nine of every 10 machines entering the TOP500, NVIDIA said, are built on NVIDIA technologies, a pace it described as compounding with each update to the twice-yearly lists.

NVIDIA tied that adoption to a broader shift toward accelerated computing, where GPUs and specialized interconnects handle the most demanding parts of workloads, rather than relying on CPUs alone. NVIDIA said its systems across the TOP500 deliver more than 2 times the AI training and nearly 3 times the AI inference throughput of every other platform combined.

Within the hardware breakdown, NVIDIA reported new records for both accelerators and interconnect. It said NVIDIA GPUs accelerated a record 238 systems on the TOP500, while NVIDIA networking connected a record 376 systems, most of them using NVIDIA Quantum InfiniBand, with the remainder using Ethernet.

The company’s push into CPUs, though still described as part of a broader “full system” approach, also featured prominently. NVIDIA said 26 systems on the TOP500 now adopt the NVIDIA Grace CPU, up eight from the previous ranking cycle. NVIDIA added that it has shipped nearly 2.5 million Grace CPUs, citing the scale of deployments it says is underway.

NVIDIA also highlighted a design choice it says helps with memory-intensive AI and scientific codes. It described the NVIDIA Grace Hopper Superchip as pairing an NVIDIA GPU with the Grace CPU in a single superchip so that the two components can share memory with minimal overhead, aiming to reduce friction between compute and data movement.

Two flagship systems served as examples of that approach in the top positions. NVIDIA said JUPITER, an NVIDIA Grace Hopper system at the Jülich Supercomputing Centre in Germany, is Europe’s fastest supercomputer and reached exascale, while also adding that the system is being used for tasks including mapping the human brain at cellular scale and simulating Earth’s climate. NVIDIA also cited Alps at No. 10 on the TOP500 and KAIROS at No. 1 on the Green500.

NVIDIA’s blog connected the rankings to the company’s energy-efficiency claims as well. It said it swept the Green500, with the top eight most energy-efficient supercomputers all running on NVIDIA GPUs, and nine of the top 10 using NVIDIA technologies. Leading the list, NVIDIA said, is KAIROS at the University of Toulouse in France, posting 73.3 gigaflops per watt, with Grace Hopper systems taking the top four spots across France, Germany and the U.K.

Still, some details remain outside the company’s disclosure. While NVIDIA quantifies system counts, throughput comparisons and adoption figures, the post does not provide a linkable methodology for the “more than 2x” and “nearly 3x” AI performance claims, nor does it break down how much of each result comes from GPUs, networking or CPU design choices. Analysts and system operators typically also factor in software stack performance, workload selection and benchmark configuration, variables the post does not enumerate.

For what comes next, NVIDIA pointed to hardware transitions tied to the newest GPU architecture cycle. The company said systems based on the Blackwell architecture, including B200 and GB200, are entering the TOP500 across Asia, Europe and the United States, with first GB200 systems debuting in Japan. It also said 35 NVIDIA AI HPC supercomputers are in development across Europe, and that new deployments span regions including South Africa, Saudi Arabia, Singapore and Vietnam. The next TOP500 and Green500 updates will test whether NVIDIA’s dominance in both speed and energy efficiency continues to widen as more Blackwell-based systems come online.