China has officially reclaimed the crown for the world’s fastest supercomputer.

A previously unlisted Chinese supercomputer, LineShine, has seized the top spot, dethroning the United States’ premier system, El Capitan, according to the latest TOP500 global rankings. Located at the National Supercomputing Centre in Shenzhen, LineShine achieved an unprecedented 2.2 exaflops (quintillion calculations per second) of sustained performance, outpacing El Capitan’s 1.8 exaflops by more than 20%.

The milestone marks China’s first appearance at the top of the list since 2017. More significantly, it represents a geopolitical triumph for Beijing, which has faced stringent U.S. export controls aimed at crippling its high-tech sector.

Unlike most modern high-end supercomputers, which rely heavily on specialized GPUs manufactured by American giants like NVIDIA Corp. and AMD, LineShine was built entirely without them. Instead, the system utilizes a unique architecture relying solely on central processing units (CPUs) — specifically, custom 304-core processors based on technology licensed from the British firm Arm Holdings.

By utilizing standard microprocessors rather than restricted artificial intelligence (AI) chips, Chinese engineers successfully bypassed U.S. trade embargoes.

“They upped us by developing a system that is not reliant on GPUs,” said Dr. Jack Dongarra, an organizer of the TOP500 list and a computer science professor at the University of Tennessee, who recently inspected the machine.

“It’s not all that surprising. China has significant technology expertise and has been in the running for some time to make the fastest machine, while the U.S., and Europe, have scaled back on their efforts mostly due to budgetary constraints,” tech analyst Jack Gold said. “It’s not just about who makes the fastest chips, as many in the U.S. think. It’s also about architectures and software capabilities, and China clearly has world class capabilities in that space.”

“China also has the ability to make their own processors, even if they may not be at the leading edge like TSMC can,” Gold added. “But they are close enough and if you build the right capabilities in processors, networking, software and memory, even if you might be at an older chip process node, it can make all the difference.”

However, the architecture comes at a high environmental cost. LineShine draws a massive 42.2 megawatts of electricity — over 40% more power than El Capitan — trading energy efficiency for raw power by linking together nearly 14 million computing cores.

While the machine represents a monument to Chinese engineering, industry analysts note that it may not give Beijing the upper hand in the broader commercial AI race.

Supercomputers traditionally excel in high-precision, 64-bit mathematics used for traditional scientific simulations. LineShine is already being deployed for complex projects, including a sophisticated Earth-system climate model and a simulation of the human brain.

Yet, when measured by the lower-precision calculations optimized for commercial AI models like those run by OpenAI or Google, LineShine ranked only fourth globally.

“China has built a magnificent machine for traditional science,” said David Harold, an analyst at Jon Peddie Research. “What it hasn’t done is close the gap as much in AI.”

The arrival of LineShine brings the total number of verified “exascale” systems in the world to five: three in the United States (El Capitan, Frontier, and Aurora), one in China, and one in Germany (Jupiter Booster).

China’s decision to submit LineShine’s test data surprised many experts, as Beijing has kept its advanced computing capabilities highly guarded in recent years. According to Dr. Dongarra, LineShine’s designers claimed the project was built without government funding, freeing them to seek international validation and submit the system for prestigious global science awards.

The breakthrough has already prompted calls in Washington, D.C., for tighter export restrictions. Jimmy Goodrich, a senior fellow at the University of California Institute on Global Conflict and Cooperation (IGCC) and a consultant to the RAND Corp., warned that the reliance on CPUs highlights a critical vulnerability in current trade policies. “The U.S. government should have stronger controls on the export and manufacturing of CPUs for the China market,” Goodrich said. “It is a loophole in the current regulations.”

As the tech race intensifies, the U.S. has begun pushing back. The Trump administration recently launched the Genesis Mission to boost supercomputing at national labs, alongside a new executive order aimed at accelerating quantum computing, the next frontier expected to surpass conventional supercomputers entirely.

“Many of the largest private computer clusters – especially those dedicated to AI – are rumored to be as large or larger than publicly funded efforts,” tech analyst Jack Poller said. “Recent reports place the xAI Colossus at around 200,000 GPUs, or approximately 25,000 nodes. I find it highly likely that between the major AI players, more than one has a cluster that exceeds LineShine’s node count and performance. However, I doubt that any of these for-profit clusters have any interest in taking the cluster off-line to run benchmarks.”