China's LineShine Tops TOP500 as Enterprise and Cloud Systems Crack Top 10

Week of June 29, 2026

Five Supercomputers Now Top One Exaflop, Led by China's LineShine

The June 2026 TOP500 list counts five systems above one exaflop on the HPL benchmark, with a CPU-only Chinese machine taking first place.

View the data

	HPL performance
LineShine (China)	2.2 EF
El Capitan (US)	1.81 EF
Frontier (US)	1.35 EF
Aurora (US)	1.01 EF
JUPITER Booster (Germany)	1 EF

More charts in By the Numbers

A Chinese system topped the TOP500 list of the world's fastest supercomputers for the first time since 2017, according to the 67th edition of the list released at ISC 2026 in Hamburg on June 29. The new No. 1, LineShine at the National Supercomputing Centre in Shenzhen, posted 2.198 exaflops on the benchmark, StorageNewsletter reported. For enterprise leaders, the bigger story sits just below the top: two systems owned by an energy company and one run by a cloud provider now rank in the global top 10, and the efficiency leaders are all built on AI-oriented hardware.

What changed at the top of the TOP500?

LineShine displaced El Capitan at Lawrence Livermore National Laboratory, which fell to second at 1.809 exaflops. Frontier at Oak Ridge (1.353 exaflops), Aurora at Argonne (1.012 exaflops) and JUPITER Booster at Jülich in Germany (1.000 exaflops) round out the five systems now above one exaflop on the High-Performance Linpack (HPL) benchmark. JUPITER remains Europe's only exascale system.

LineShine is unusual. According to the report, it is a CPU-only machine with about 13.79 million cores, draws roughly 42.2 megawatts, and reached about 80% of its theoretical peak of 2.736 exaflops. It has no dedicated low-precision accelerators, the kind of hardware that powers most modern AI training and inference.

That design shows up in the mixed-precision results. On HPL-MxP, a benchmark closer to AI-style math, El Capitan leads at 16.7 exaflops, or 9.2 times its standard HPL score. Aurora (11.6 exaflops) and Frontier (11.4 exaflops) follow, with LineShine fourth at 7.92 exaflops. LineShine did take first place on the HPCG benchmark, which stresses memory and communication patterns common in real applications, at 22.00 petaflops.

Why do enterprises appear in the top 10?

Italian energy company Eni has two systems in the top 10. Its new HPC7 debuted at No. 6 with 571.5 petaflops, using the same HPE Cray EX255a and AMD MI300A architecture as El Capitan, while its HPC6 sits at No. 8. Microsoft Azure's Eagle ranks seventh at 561.2 petaflops.

Those placements reflect a shift AI Connect speakers have described for some time: the line between scientific computing and enterprise AI is blurring. "Really this has become a supercomputing problem," HPE Federal AI Chief Technologist Steve Heibein said in a AI Connect webinar on how AI and HPC simulation are converging, describing how simulation teams and AI teams now need the same large, dense clusters.

Vendor concentration is also worth noting. HPE supplied six of the top 10 systems. AMD powers four and accounts for more than 40% of top-10 HPL performance, according to the report. Nvidia hardware appears in three top-10 systems: Eagle, JUPITER and Alps.

What does the Green500 say about AI efficiency?

The efficiency rankings point toward the hardware enterprises are buying for AI. The top three Green500 systems, KAIROS at the University of Toulouse (73.28 gigaflops per watt), ROMEO-2025 in France (70.91) and the Levante GPU extension at DKRZ in Germany (69.43), all use BullSequana XH3000 systems with Nvidia Grace Hopper Superchips. The report notes smaller installations tend to rank higher on efficiency.

For comparison, El Capitan delivers 60.94 gigaflops per watt on HPL, while LineShine manages 52.07. Power is the binding constraint for most corporate AI build-outs, and Heibein put it plainly in the same webinar: "A thousand watt GPU needs a thousand watts of cooling."

The mixed-precision results carry another enterprise signal. SoftBank's CHIE-4 posted the largest gain of any system, with an HPL-MxP score 24.4 times its standard result, a sign of how heavily AI-oriented systems lean on lower-precision math.

What should enterprise leaders take from the list?

The TOP500 is a benchmark contest, not a purchasing guide, but it offers a few useful reads for CIOs and infrastructure teams:

• Industrial AI infrastructure is reaching national-lab scale. Eni's two systems show that companies with heavy simulation and AI workloads are building at a scale once reserved for governments.
• Measure what matters for your workloads. The gap between HPL, HPCG and HPL-MxP rankings shows that the "fastest" system depends on the work. The same holds for enterprise AI clusters, where inference latency and cost per task matter more than peak numbers.
• Efficiency and sovereignty are rising priorities. A CPU-only Chinese system at No. 1 underlines how geopolitics shapes access to accelerators. Our coverage of sovereign AI with NVIDIA and HPE and of enterprise AI factories explores how organizations weigh control against capability.

The list is updated twice a year. Between editions, the things to watch are whether more corporate systems climb the rankings, and whether efficiency gains on AI-oriented hardware keep outpacing growth in raw performance.

Frequently asked questions

What is the fastest supercomputer in June 2026?

LineShine at the National Supercomputing Centre in Shenzhen, China, ranked No. 1 on the 67th TOP500 list with 2.198 exaflops on the HPL benchmark.

How many exascale supercomputers are there?

Five systems exceed one exaflop on HPL: LineShine, El Capitan, Frontier, Aurora and JUPITER Booster, which is Europe's only exascale system.

Which companies have systems in the TOP500 top 10?

Energy company Eni has HPC7 at No. 6 and HPC6 at No. 8, and Microsoft Azure's Eagle ranks No. 7.

Filed under: Data & AI Infrastructure