INVESTIGATION: LineShine — How China Reclaimed the Top of Global Computing Despite Sanctions

INVESTIGATION: LineShine — How China Reclaimed the Top of Global Computing Despite Sanctions

Introduction: On June 24, 2026, Beijing takes the lead in global supercomputing

TOP500, 67th edition: a ranking that changes everything

On June 24, 2026, the 67th edition of the TOP500 ranking — the biannual list of the 500 most powerful computing systems in the world, the sector's absolute benchmark — delivered a verdict that sent shockwaves through Western technology intelligence circles: a Chinese supercomputer, LineShine, installed at the National Supercomputing Center of Shenzhen (NSCS), seized first place globally with a performance of 2.198 exaflops. This is the first time since 2017 that a Chinese system has led this ranking. And it is the first time in history that a supercomputer has exceeded two exaflops using only conventional processors (CPU) — without a single American graphics processing unit (GPU).

The significance of this event reaches far beyond the world of specialized computing professionals. LineShine is built on LX2 processors developed in China, a proprietary architecture called "LingKun," a proprietary interconnect "LingQi," and a Kylin OS operating system. Not a single American component in a machine that has beaten the best achievements of AMD, Intel, and their partners. This is a demonstration of computational independence that Washington spent years trying to prevent through export control policies — and this demonstration has just proven that policy failed in its primary objective.

What 2.198 exaflops actually means

2.198 exaflops — or 2.198 quintillion floating-point operations per second. To grasp this power: this figure represents approximately 2,200,000,000,000,000,000 mathematical operations per second. If every human being on Earth performed one mathematical operation per second, it would take 280 million years to do what LineShine does in one second. This is computing power that transforms the possibilities of research in physics, meteorology, molecular biology, artificial intelligence, cryptography — and weapons development.

The previous world number one, the American El Capitan at Lawrence Livermore National Laboratory, reaches 1.809 exaflops — about 21% less powerful than LineShine. Frontier (Oak Ridge National Laboratory) is third with 1.353 exaflops, Aurora (Argonne National Laboratory) fourth with 1.012 exaflops. China has not merely caught up with the United States in supercomputing — it has surpassed it.

LineShine's architecture: a calculated technological independence

The LX2 processor: the chip that changed everything

At the core of LineShine sits the LX2 processor — a chip with 304 cores, running at 1.55 GHz, developed in China within the "LingKun" platform. This processor is the product of a decade of massive investment in Chinese semiconductor research, catalyzed by — in history's irony — the American sanctions themselves. When Washington cut China's access to advanced chips from NVIDIA, AMD, and Intel, it forced Chinese engineers and institutions to develop their own alternatives. LineShine is the product of this constraint transformed into opportunity.

The complete system mobilizes 13.79 million cores — a staggering figure representing a computing infrastructure of incomprehensible scale to the general public. To power this infrastructure, LineShine consumes approximately 42.2 megawatts of electrical power. For comparison: this is the electrical consumption of approximately 35,000 European households. This is not a machine — it is a city of computation.

LingQi: the proprietary interconnect that makes the difference

In supercomputers, performance depends not only on the power of individual processors — it depends on the speed at which those processors can communicate with each other. This is the role of the interconnect — the system's internal network. LineShine uses a proprietary interconnect called "LingQi," developed in China, which ensures data transfers at sufficient speed to coordinate its 13.79 million cores in real time.

This proprietary interconnect is, along with the LX2 processor, the second pillar of LineShine's technological independence. The highest-performing interconnects in American systems use InfiniBand technology (NVIDIA network) — a technology China can no longer freely procure under export restrictions. Developing a proprietary interconnect of equivalent quality is one of the most significant technical achievements of this project.

Secondary rankings: a more nuanced picture

HPCG and HPL-MxP: where LineShine wins and where it falls behind

The TOP500 ranking uses several different metrics to evaluate supercomputers. The HPL (High Performance Linpack) — on which LineShine ranks number one — measures performance in dense matrix computation, representative of traditional scientific applications. The HPCG (High Performance Conjugate Gradients) measures a more complex workload closer to real-world engineering applications. The HPL-MxP measures mixed-precision performance, crucial for AI workloads.

On HPCG, LineShine is also number one with 22.00 HPCG-Petaflop/s, ahead of El Capitan (17.41). This is a double victory. On HPL-MxP — the ranking crucial for artificial intelligence and language model training — LineShine falls to fourth place with 7.92 exaflops, behind El Capitan (16.7), Aurora (11.6), and Frontier (11.4). This deficit is explained by LineShine's CPU-only architecture: GPUs are optimized for the mixed-precision computations used by AI — and the most powerful GPUs remain American.

What the HPL-MxP gap means for Chinese AI

LineShine's fourth place in HPL-MxP is important information. It means that for applications training large AI models — those requiring massive quantities of mixed-precision computation — China has not yet achieved parity with the best American systems. El Capitan delivers 16.7 exaflops in HPL-MxP versus 7.92 for LineShine — more than double.

This AI dimension deficit is a real strategic vulnerability for China. In the race for artificial intelligence — which may be the determining factor of twenty-first-century technological competition — NVIDIA GPUs and their equivalents remain dominant. American export sanctions on advanced GPU exports to China (H100, A100, and successors) create a real constraint that LineShine does not overcome. For AI, the Sino-American technological battle is not yet decided.

The sanctions strategy: a policy in partial failure

How Washington tried to block China

The American policy of technology export controls toward China has a long history. It was intensified under the Biden administration with restrictions in 2022 and 2023 on advanced chip exports — prohibiting notably the sale to China of NVIDIA's H100 and A100 GPUs, the most advanced chips commercially available. These restrictions explicitly aimed to deprive China of the components needed to build supercomputers and train top-tier AI models.

The results of this policy are now legible in the TOP500 ranking. On one hand, the restrictions clearly slowed Chinese access to the most advanced GPUs — as evidenced by LineShine's lag in HPL-MxP. On the other, these same restrictions catalyzed the development of China's domestic semiconductor industry, producing the LX2 processor and the LingKun architecture. This is the Sputnik effect in semiconductor form: pressure creates innovation.

The technology sanctions paradox

The history of technology sanctions shows a recurring paradox. In the short term, they effectively slow the targeted country — China would probably not have built LineShine as quickly had American GPUs been available. In the medium term, they force the targeted country to develop its own capabilities — as China did, as the USSR had developed its atomic bomb after the United States tried to maintain that monopoly, as Japan developed its consumer electronics after American embargoes on raw materials in the 1930s.

The question that LineShine poses to Washington is blunt: did technology sanctions achieve their objective — maintaining American technological superiority over China in supercomputing? The answer is no. They created a delay. They may have prevented even faster development. But they did not prevent China from building the world's most powerful supercomputer.

The strategic stakes: supercomputing as national power infrastructure

What is a 2-exaflop supercomputer actually used for?

Why is a supercomputer like LineShine strategically important? Because computing power directly determines a country's capability in critical domains. In climate research: high-resolution climate models require exaflops of computation. In pharmaceutical development: drug design through molecular simulation requires massive computation. In nuclear physics: simulation of thermonuclear explosions — which allows nuclear arsenals to be maintained without physical tests — requires supercomputers of the El Capitan or LineShine generation.

In artificial intelligence: even if GPUs remain dominant for training, supercomputers play a growing role in scientific AI applications. In cryptography: breaking current encryption systems requires massive computing power, and the race between encryption and decryption is partly played out on supercomputers. And in weapons development: missile aerodynamics, war scenario simulation, trajectory optimization — all direct military applications.

Nuclear weapons simulation: the taboo that must be named

Lawrence Livermore National Laboratory, which houses El Capitan, has a primary mission: simulation of American nuclear weapons — ensuring that the arsenal remains reliable and functional without physical testing, which has been prohibited since the Comprehensive Nuclear Test Ban Treaty. The fact that El Capitan is now second behind LineShine raises a question that nobody poses openly but that everyone is thinking: if China uses LineShine to simulate its nuclear weapons, it now possesses potentially superior simulation capability to that of the United States.

I cannot confirm that China uses LineShine for military or nuclear applications — publicly available information does not say this, and it would be irresponsible to assert it without evidence. But the hypothesis is not absurd: the world's most powerful supercomputers almost always serve, in part, military applications. This is a reality that strategic assessment of LineShine must take into account.

The American response: El Capitan and the GPU ecosystem

El Capitan: the American response that remains competitive

El Capitan, with its 1.809 exaflops in HPL, remains a machine of extraordinary power — and in HPL-MxP, it still largely dominates LineShine with 16.7 versus 7.92 exaflops. This AI computation superiority results from El Capitan's hybrid architecture: it combines AMD EPYC 4th-generation processors with AMD Instinct MI300A accelerators — high-performance GPUs specialized for AI workloads.

This HPL-MxP superiority reflects the American structural advantage in AI GPUs. NVIDIA and AMD remain global leaders in advanced computing GPUs — and their most powerful products are not available to China under export restrictions. For generative AI workloads, language model training, and mixed-precision computation, American systems maintain a clear lead.

The American innovation cycle: Aurora, Frontier, and beyond

The American supercomputing landscape does not reduce to El Capitan. Aurora (1.012 exaflops) at Argonne uses Intel Ponte Vecchio GPUs. Frontier (1.353 exaflops) at Oak Ridge relies on AMD GPUs. Together, these three systems represent a depth of computing capability that China, with a single public exascale system, cannot yet match in aggregate.

Moreover, the United States already has next-generation supercomputer projects in development — "post-exascale" systems targeting 10 exaflops or more in the years ahead. If development stays on schedule — never guaranteed in projects of this complexity — the United States could reclaim first place within two to three years. But China is also developing its next-generation systems. The competition will continue to intensify.

Chinese computational independence: a political project as much as a technical one

Kylin OS: even the operating system is Chinese

LineShine runs Kylin OS — an operating system developed in China, based on the open-source Linux kernel but deeply customized by Chinese teams. This choice reflects the doctrine of integral computational independence guiding LineShine's development: no dependence on any foreign technology at critical system layers.

This doctrine is not merely technical — it is political. It responds to the very real fear that foreign-sourced software or hardware might contain backdoors, secret access points, or remote sabotage mechanisms — a fear the United States itself used to justify excluding Huawei from its 5G networks. China is applying at the national level the same digital sovereignty logic the West applies to its critical infrastructure.

The "Made in China 2025" plan and its fruits

LineShine is the visible fruit of an industrial policy launched as early as 2015 under the "Made in China 2025" plan — a government program aimed at making China independent in critical technology sectors. This plan explicitly targeted semiconductors, supercomputers, AI, aeronautics, and biotechnology. It was funded by tens of billions of dollars in public investment and mobilized the best Chinese universities and research centers.

Western critics tended to dismiss this plan as industrial propaganda — political ambition without real technical results. LineShine refutes this reading. The "Made in China 2025" plan has delivered results — not in all sectors, not at the announced pace — but in supercomputing, the result is there, measured and ranked by an independent international institution. It is a documented victory of a long-term state industrial policy.

Implications for the global AI race

Supercomputing and AI: two faces of the same competition

The supercomputing race and the artificial intelligence race are inseparable. Training a large language model (LLM) like GPT-4 or its successors requires hundreds of thousands of GPUs running in parallel for weeks. The data centers of Microsoft, Google, Amazon, and Meta represent concentrations of computing power comparable to supercomputers — often measured in aggregate exaflops.

China has developed its own LLMs — DeepSeek, ERNIE, Baidu models — with performance comparable to the best American models on certain tasks. But this development runs up against the restriction of advanced GPUs, which limits the size and training speed of models. If China succeeds in developing domestic GPUs equivalent to NVIDIA's H100 or B200 chips — a race in which it is investing massively — the current AI constraint could disappear within five years.

DeepSeek as an early warning signal

The release of DeepSeek in early 2025 had already sent an alert signal to the American AI ecosystem: a Chinese model developed with less advanced GPUs and at notably lower cost had achieved performance comparable to the most advanced American models. This was not a total victory — the best American models remained superior on certain complex tasks. But it was the demonstration that algorithmic ingenuity can partially compensate for limited hardware access.

DeepSeek + LineShine = a Chinese AI ecosystem advancing on all fronts simultaneously. The algorithms improve. The computing infrastructure intensifies. Export restrictions create delays but not permanent blockages. The AI race with China is closer than most Westerners believe.

Shenzhen, NSCS, and China's innovation ecosystem

Why Shenzhen?

The installation of LineShine at the National Supercomputing Center of Shenzhen (NSCS) is not accidental. Shenzhen is China's Silicon Valley — a metropolis of 17 million people built from scratch since the 1980s, which today houses Huawei, Tencent, BYD, DJI, and thousands of startups and R&D laboratories. It is the city that best symbolizes China's technological transformation — a transformation planned by the state but fueled by genuine entrepreneurial and engineering energy.

Installing the world's most powerful supercomputer in Shenzhen sends a double message: domestically, it validates the technological development model of the laboratory city. Externally, it positions Shenzhen — and China — as the new global center of advanced computing. This geopolitical signal is intentional and meticulously calculated.

University-industry-state cooperation

The development of LineShine illustrates the Chinese model of "civil-military fusion" — close collaboration between university research institutes, private companies, and the state. The Shenzhen Cloud Computing Center that built the system, the engineering teams of the LingKun platform, and the government funds that financed the project — all of this operates in an ecosystem where boundaries between public and private sectors are deliberately blurred and mutually beneficial.

This model is different — and in certain respects more effective for large-scale national projects — than the American model, where supercomputers are primarily developed by national laboratories funded by the Department of Energy. The comparison is not simple: each model has its strengths. But China's ability to rapidly mobilize massive resources for priority technology projects is documented by LineShine.

Global reactions: from shock to strategic reassessment

The American reaction: a strategic pause

The official American reaction to the TOP500 result has been measured — publicly. In private, according to multiple sources close to defense and technology intelligence circles, the result has triggered an urgent revision of assessments of Chinese technological capabilities. Intelligence agencies, national laboratories, and the Department of Defense are all reassessing their assumptions about the pace and scale of Chinese computing development.

A specific question has surfaced in these discussions: is LineShine the only system of this class, or is it the public version of an even more powerful Chinese military computing infrastructure? This is a question that open sources cannot answer — but it illustrates the strategic anxiety that the TOP500 ranking generates well beyond the circle of computing specialists.

The European reaction: between concern and missed opportunity

Europe watches LineShine's success with particular anxiety. While the United States and China both have exascale systems, Europe remains behind in this race. The most powerful European supercomputer — LUMI in Finland — is of a previous generation and sits at a fraction of the power of LineShine or El Capitan. The European High Performance Computing initiative (EuroHPC) continues its development program, but the pace and ambitions remain below those of the two supercomputing superpowers.

For Europe, LineShine's success should be an alarm bell: in the race for advanced computing, standing still means falling behind. Computing capabilities directly determine competitiveness in AI, pharmaceutical research, climate simulation — all domains in which Europe has vital interests. Remaining in the second tier of global supercomputing is a mortgage on European technological sovereignty.

Toward the next generation: who will be number one in eighteen months?

American post-exascale systems in development

Washington is not standing still. Several next-generation supercomputer projects are under development in the United States, with target performances ranging from 5 to 20 exaflops. These systems will use next-generation GPUs from NVIDIA (Blackwell and successors) and AMD, and will benefit from advances in interconnects and energy efficiency. If timelines hold — never guaranteed in projects of this complexity — the United States could reclaim first place within two to three years.

But China is also developing its next-generation systems. The engineers of LingKun and LineShine are not stopping after their victory — they are already preparing the next generation of LX processors and computing architectures. And if American export restrictions continue to drive the development of Chinese domestic GPUs, the current HPL-MxP constraint could be lifted within five years.

A world of five exascale systems: a new normal

The TOP500 report of June 2026 notes a historic change: for the first time, there are simultaneously exascale systems in Asia, North America, and Europe. There are now five systems exceeding one exaflop in HPL. This "globalization of supercomputing" means that the American monopoly on cutting-edge extreme computation — which had lasted several years — belongs to the past.

This is not necessarily bad news for global science — more exascale systems, geographically distributed, means more capacity for climate, medical, and physics research. But from the perspective of strategic competition, it means that technological superiority can no longer simply rest on access to computing infrastructure that adversaries lack. The competition shifts to other fronts: algorithms, applications, talent management, systems security.

Lessons for the West: invest or fall behind

The Western underinvestment deficit in supercomputing

The West — taken as a whole — is underinvesting in supercomputing relative to its technological ambitions. The United States maintains leadership through its national laboratories, but budgets are constrained and development timelines long. Europe is structurally behind. Japan, with its Fugaku system, remains competitive but lacks portfolio depth.

The lesson of LineShine is simple: technological superiority does not maintain itself. It requires continuous, ambitious, and strategically coordinated investment. Export restrictions can slow adversaries, but they do not replace indigenous innovation. If the West wants to maintain its superiority in advanced computing — and in the AI that depends on it — it must invest massively, coordinate its efforts (especially in Europe), and develop the human talent that feeds these systems.

The talent battle: the other dimension of the race

Behind every supercomputer, there are engineers. China trains hundreds of thousands of engineers annually in computing, electronics, and applied mathematics. Many of them were trained in American and European universities — and a policy of restricting student visas, as some American political circles have considered, could deprive the West of precious talent while reinforcing Chinese capabilities at home.

The real answer to LineShine is not an additional technology restriction law — it is investing in scientific education, attracting and retaining the world's best engineers (including Chinese engineers, who contribute massively to American innovation), funding ambitious computing projects, and maintaining transatlantic research partnerships. This is not naivety — it is the policy that produced the American era of supercomputing in the 2010s.

The impact on military AI: supercomputing and the AI arms race

Supercomputing as military AI infrastructure

LineShine's victory in the TOP500 ranking is not merely an academic or industrial achievement — it has direct implications for global military competition. Supercomputing is the foundational infrastructure of military artificial intelligence: missile guidance systems, real-time intelligence analysis, tactical scenario simulation, development of new hypersonic weapons. A nation that masters supercomputing holds a decisive advantage in developing these capabilities.

The Chinese military — the People's Liberation Army — is investing massively in integrating AI into its military operations. China has publicly announced its intention to achieve military parity with the United States by 2035. The availability of world-class domestic computing power, beyond the reach of American sanctions, is a key element of this ambition. LineShine is not merely a civilian supercomputer — it is a dual-use infrastructure whose military beneficiaries are as real as its scientific ones.

The algorithmic arms race: a new front of competition

The war in Ukraine demonstrated that algorithms have become weapons in their own right: target recognition, drone trajectory optimization, electronic intelligence analysis. These algorithms require considerable computational resources for their training and deployment. The nation with the greatest computing power holds a potentially decisive advantage in developing these algorithmic weapons.

For NATO member states, the lesson of LineShine is therefore twofold: Beijing has crossed a computational threshold that gives it an accelerated capacity to develop AI-based military capabilities. The Atlantic Alliance's commitments to investment in military AI and supercomputing — still insufficient — must be urgently reinforced if the West is to maintain its operational advantage in the coming decade.

The talent ecosystem: where are tomorrow's supercomputing engineers trained?

Engineer training: a demographic and educational competition

Behind supercomputers, there are engineers. Each year, China trains hundreds of thousands of engineers in computing, electronics, and applied mathematics — several times more than the United States and Europe combined. Not all of these graduates work in supercomputing. But the depth of the talent pool available for projects like LineShine is unmatched in the Western world. And this pool grows denser every year.

China has also developed a very effective talent repatriation system: incentive programs attract Chinese researchers trained abroad — at the best American and European universities — to return to work in China. The "Thousand Talents" program, despite the political controversies it has generated, has considerably enriched the Chinese scientific ecosystem. This transfer of Western expertise to China is a factor often underestimated in Beijing's technological rise.

The West facing a talent deficit in advanced computing

American and European universities also train world-class engineers and researchers in supercomputing and advanced computing. But their numbers are insufficient relative to sector needs. Demand for talent in AI, supercomputing, and cybersecurity far exceeds the available supply. And a portion of the best talent — particularly those of Asian or international origin — chooses to return to their home countries or is actively recruited by Chinese companies.

The Western response must include a strong educational dimension: drastically increasing the number of advanced computing graduates, strengthening the appeal of careers in public technology, maintaining openness to international talent while protecting sensitive strategic knowledge. This is not a contradiction — it is the sophisticated management of a competition that is also played out in university lecture halls.

Conclusion: LineShine as warning and as challenge

What LineShine says about the world in 2026

LineShine says several things simultaneously about the world in 2026. It says that China has achieved technological maturity in supercomputing — a domain that was exclusively Western a decade ago. It says that technology sanctions cut both ways. It says that state technological ambitions, sustained over decades, can produce results that short-term democracies underestimate. And it says that the Sino-American technological competition — which is at bottom the principal geopolitical fracture of the twenty-first century — has intensified, not diminished.

This is not a catastrophe. The West has the resources, talent, and institutions to respond. But the response must match the stakes — not press releases about export restrictions, but massive, sustained, coordinated investment in the technological infrastructure that will determine tomorrow's balance of power.

LineShine's victory: permanent or temporary?

LineShine's number-one position is permanent in the moment — on the June 2026 ranking, it is set. It is temporary in the ongoing competition — in six months, in a year, in two years, the ranking will evolve. The United States is developing more powerful successors to El Capitan. China is developing successors to LineShine. The race does not end.

What matters is what LineShine represents symbolically and strategically: a China technologically independent in advanced computing, capable of building the world's best system with its own chips, its own network, its own OS. This computational independence is irreversible. And it fundamentally transforms the terms of global technological competition.

By Maxime Marquette, columnist

Columnist's transparency note

Skills and limits

I am Maxime Marquette, analyst-columnist in geopolitics and technology. I am not a computing engineer or semiconductor specialist. My analysis of LineShine is based on primary public sources — the TOP500 website, The Guardian, Washington Post, Nature — and on my general knowledge of technology strategy. For purely technical aspects (processor architecture, specific benchmarks), I defer to data published by TOP500 and analyses by cited specialists.

Assumed biases

I am concerned by the rise of Chinese technological power under the authoritarian regime of the Chinese Communist Party. I believe the West must invest massively in its own technological capabilities rather than focusing exclusively on restrictions. These positions influence my framing — I assume and explain them.

Sources

Primary sources

TOP500.org — LineShine debuts at No. 1, TOP500 enters new global exascale era — June 24, 2026

The Guardian — Chinese supercomputer overtakes best US machines to become world's fastest — June 24, 2026

Nature — China's LineShine just topped the global supercomputer rankings — what you need to know — June 26, 2026

Secondary sources

Washington Post — Chinese supercomputer dislodges US machines as world's fastest for first time since 2017 — June 23, 2026

Tech Wire Asia — China beats US in TOP500 ranking with world's fastest supercomputer — June 24, 2026

The Register — Chinese supercomputer using local processors heads TOP500 list — June 24, 2026