The most powerful supercomputer in the world is in China. The other part of the story is even more interesting

The biannual TOP500 list with the most powerful supercomputers on the planet has given a striking surprise in its June 2026 edition. The Chinese LineShine system, installed at the Shenzhen National Supercomputing Center, has debuted directly at number one. It thus displaces the American supercomputer El Capitan, which had dominated the ranking for years. China has not managed to lead this classification since 2017, thus breaking a decade of North American hegemony.

Unprecedented raw power. The performance tests used to configure this list leave no doubt: LimeShine has achieved 2,198 exaflops of performance in the benchmark HPLcompared to 1,809 exaflops for its American rival. The Chinese machine is therefore 20% more powerful than the flagship of the Lawrence Livermore National Laboratory in California. It is a whole new milestone in global supercomputing.

Surprise: zero GPUs. The performance is extraordinary, but even more so is the way this supercomputer has been created. Most modern supercomputers rely heavily on GPUs, specialized graphics chips from Nvidia or AMD, for massive data processing. However, LineShine uses CPUs instead of focusing everything on GPUs, something that differentiates this supercomputer from its rivals and makes the feat even more striking.

Heart ARM. The fundamental pillar of LineShine It is the LX2 CPU. The data they point because it has been designed by Huawei, and in each of those CPUs we have two dies HBM computing and memory. Each die It has 152 ARMv9 cores that have EVS and EMS supportwhich allows the system to process vectors and matrices in an exceptional way even without GPUs. In total LineShine has 304 of these processors with a total of 13,789,440 cores.

Avoiding vetoes. One of the reasons that have undoubtedly contributed to this design decision is the US trade war with China. Tariffs and export bans on hardware and software have made things very complicated, especially when it comes to getting Nvidia GPUs for AI processing. Despite all this, China has once again demonstrated an astonishing ability to advance technologically. Another curiosity: the system has been built without public funds from the Chinese government.

Source: TOP500.org

AI clusters don’t compete here (but they would “win”). This prestigious list had always offered us that vision of the most powerful computing systems in the world, but today the panorama has changed. It has done so because the AI ​​clusters created by big technology are probably more powerful than any of these systems. As explains Jimmy Goodrich of the University of California, “if hyperscalers competed with their systems, ‘fastest’ in the world wouldn’t even be in the top five.” That phrase, yes, has crumbs.

But it’s comparing pears with apples. However, the supercomputers on the TOP500 list and the AI ​​clusters that hyperscalers are building to meet global demand are very different. The root problem is in floating point precision. Classic supercomputers like El Capitan are designed for high-fidelity scientific simulations, where the slightest rounding error can be fatal. That is why they operate under the FP64 standard with which tens of decimals are calculated: it is a slow and expensive process in energy, but extraordinarily precise.

AI rounds with joy. In contrast, AI models are very resistant to numerical “noise.” They don’t need perfect precision to recognize patterns or generate text. That allows AI chips to use reduced precision formats like FP16, FP8, or even FP4. By processing much shorter numbers, they multiply their speed and efficiency significantly. So when an AI cluster achieves tens of exaflops, it does so rounding up quite happily. These machines are exceptional for AI tasks, but they do not threaten the future of traditional supercomputers.

Europe (and Italy) and supercomputing. If we look in detail at the list, we see great news for European supercomputing. The HPC7 system created by Eni has entered directly at number 6 on the list, while Jupiter Booster (Germany) is at number 5. Europe has four systems in the top 10 of the TOP500 list (two of them, from Italy), and eight in the top 20. Spain is still present on the list thanks to MareNostrum 5which yes, drops from 14th to 16th place.

In Xataka | The EU wants to close the gap in the race for AI with 750 million euros. And it is good news for Barcelona