ANALYSIS: Why China Has Regained the Title of the World’s Most Powerful Supercomputer

Maxime Marquette

2026-06-26 22:13:57

CPU-only architecture: the choice that changes everything

LineShine is built entirely on CPUs—central processing units—without any GPUs (graphics processing units). This represents a radical departure from the philosophy that has dominated global supercomputing for the past decade. Virtually all of the machines at the top of the TOP500 list—including El Capitan—combine CPUs with GPU accelerators for massively parallel tasks. Nvidia GPUs dominate this ecosystem. And specifically, high-performance Nvidia GPUs have been subject to U.S. export controls to China since 2022.

Beijing has therefore circumvented the obstacle by eliminating it: LineShine runs on the LX-2 processor, designed in China and based on the ARMv9 architecture, with more than 300 compute cores per chip. The entire system incorporates a total of approximately 13.789 million cores—across 45,360 LX-2 processors clocked at 1.55 GHz, running the Kylin OS (a Chinese version of Linux). As Jack Dongarra himself notes, this is the first time a computer using only CPUs has reached the exaflop scale. A milestone in the history of computing.

Shenzhen: A Symbol of China’s Technological Rise

Shenzhen is not a random choice. This city in southern China, which was a fishing village forty years ago, has become China’s Silicon Valley—home to Huawei, BYD, DJI, and countless tech startups and electronics manufacturers. Installing the world’s most powerful supercomputer in Shenzhen is a statement of geographic intent: this is where China’s 21st-century technological dominance is being built.

The Shenzhen National Supercomputing Center (NSCS) is one of China’s six national supercomputing facilities. Unlike U.S. centers such as Lawrence Livermore or Oak Ridge—whose mandates are largely military and nuclear—the NSCS in Shenzhen has an explicitly commercial and civilian purpose. It serves as a platform for industry, pharmaceutical research, climate simulations, and training artificial intelligence models. The line between civilian and military use is obviously more blurred in China than it appears, but the public image matters.

With Shenzhen embodying technology, Beijing politics, and Shanghai finance—China has orchestrated its rise with a coherence that the West struggles to replicate. Our democracies spend ten years debating the location of a semiconductor factory. China builds the world’s most powerful supercomputer and announces it to the entire world in Hamburg. There’s a lesson in that one that we’re reluctant to learn.

Why U.S. Export Controls Failed

The Logic Behind Technology Sanctions and Their Limitations

Since 2022, the United States has gradually tightened its embargo on advanced semiconductors exported to China. The explicit goal: to deprive Beijing of the GPUs and manufacturing equipment needed to train AI models and run state-of-the-art supercomputers. Companies such as Nvidia, ASML, and TSMC have been barred from selling their most advanced technologies to Chinese customers. In theory, this amounts to a technological stranglehold. In practice, LineShine demonstrates the limits of this strategy.

The reality is that China has responded to the sanctions with a strategy of forced innovation. Deprived of Nvidia GPUs, it has invested heavily in developing its own chips. The LX-2 processor is the result of this policy. It does not outperform Nvidia’s best GPUs in terms of energy efficiency or performance per chip—El Capitan consumes power at a level that available data does not allow for a direct comparison with LineShine. But LineShine proves that by stacking enough of its own chips, China can reach the top. It is this “quantity over quality” approach that compensates for the embargo.

China’s Mysterious Disappearance from the TOP500 and What It Reveals

One troubling detail deserves attention. Since around 2022, Chinese institutions had virtually stopped submitting their systems to the TOP500 rankings. The number of Chinese submissions had dropped to just a handful, and none came close to the top. Western analysts had concluded that China was stagnating. This was a misinterpretation: China was deliberately keeping quiet, secretly developing exascale systems that it did not disclose—such as OceanLight or Tianhe-3 (“Xingyi”).

LineShine thus marks a deliberate return to the public arena. Beijing has now chosen to unveil what it has built. The timing is no coincidence: it coincides with a period of heightened trade and technology tensions with Washington. It is as much a political message as it is a technological achievement. The message: “Your sanctions haven’t stopped us. We’re here.”

For years, some Western experts took pride in the fact that China was absent from the TOP500. “They have nothing left to show,” they said. That was exactly the kind of arrogance that precedes an unpleasant surprise. China wasn’t playing by the rules of international transparency—it was building behind the scenes. LineShine is the result of that silence. It would be wise to learn from this before the next list is released in November.

Strategic Significance: Computation, AI, and Military Supremacy

What are supercomputers really used for?

You might think the race for supercomputers is all about national pride. It’s much more than that. A supercomputer of this caliber makes it possible to do things that would otherwise be impossible: simulate nuclear reactions without physical testing, model the effects of a biological weapon, train large language models for AI, simulate the behavior of new materials for military applications, predict the trajectories of hypersonic missiles, or simulate cyberwarfare scenarios.

El Capitan, the dethroned American champion, is managed by the U.S. Department of Energy and is used primarily to maintain the U.S. nuclear arsenal without physical testing. LineShine has a stated civilian purpose—but in a state-party where the civilian-military divide is a legal fiction, this distinction is purely formal. Computing power is a strategic resource on par with oil or rare earth elements.

The AI Rankings and What LineShine Doesn’t Do (Yet)

An important caveat is in order. On the HPL-AI benchmark—designed to simulate workloads similar to those of artificial intelligence—LineShine ranks fourth, far behind systems optimized for AI with Nvidia GPUs. This reflects the architectural limitation of an all-CPU machine in a world where AI is massively accelerated by GPUs. U.S. systems maintain a significant lead in this specific area.

But this caveat should not diminish the achievement. 2.198 exaflops of LINPACK performance, measured and submitted to an independent organization—that is a scientifically verified fact. Using its own chips, China has built a system that no one else in the world has been able to match in terms of raw computing power. This is enough to shake Western confidence in the ability of its sanctions to contain China’s technological rise.

LineShine isn’t yet the world’s best AI tool—and Americans will take comfort in that. But it proves that China can design and manufacture its own world-class chips. It’s not today’s AI that’s at stake here. It’s tomorrow’s capabilities. And no one can claim any longer that those capabilities are locked behind export sanctions.

Conclusion: The Lesson from LineShine for the West

A symbolic victory with concrete implications

LineShine is a demonstration of technological sovereignty. China has built the world’s most powerful supercomputer by ignoring the Western rules of the game: no Nvidia GPUs, no ASML ultra-fine lithography, no TSMC. Its own chips, its own software, its own operating system. It’s not perfect—the AI performance is inferior—but it’s functional, verified, and public. The West can no longer convince itself that its technological restrictions are enough to maintain its lead.

The appropriate response is not panic. It is acceleration. Europe and the United States must invest heavily in their own computing capabilities, their own chip architectures, and their own world-class supercomputing centers. The era when Western technological supremacy was structurally guaranteed is over. LineShine has just signed its symbolic death certificate.

The Challenge for the Coming Years

The next TOP500 list will be published in November 2026. The United States is already working on the next generation of machines—the successor to El Capitan is in development. Europe, with its EuroHPC projects, is trying to stay in the conversation. But the announcement of LineShine has set the terms of the technological competition for the coming years: China can do on its own what the West thought it could prevent. That is the key takeaway from the ISC Conference in Hamburg, June 2026. Everything else is ancient history.

By Maxime Marquette, columnist