SAN FRANCISCO — China retook, this Tuesday (23), a coveted computing position from the United States, elevating the technological dispute between the two powers in a race with implications for science, national security and geopolitics.
LineShine, a large computing system in Shenzhen, China, has been declared the fastest in the world by a group of researchers based on a series of standard tests for supercomputers. In addition to raw speed, the system drew attention for using only conventional microprocessors, and not the specialized chips known as graphics processing units, or GPUs, that most high-end supercomputers rely on to perform heavy calculations.
This basic design could point to a more efficient way of combining artificial intelligence with traditional scientific tasks, said Jack Dongarra, one of the organizers of the Top500 list, which brings together the world’s most powerful supercomputers.
Dongarra, a professor of computer science and electrical engineering at the University of Tennessee, recently inspected the new machine at the Shenzhen Cloud Computing Center. The results obtained by LineShine were more than 20% higher than those of El Capitan, a system at Lawrence Livermore National Laboratory, in California, which has led the biannual supercomputer performance ranking since November 2024. China has not placed a machine at the top of the list since 2017.
“It’s an impressive system,” Dongarra said of LineShine. “They outperformed us by developing a system that doesn’t rely on GPUs.”
The new supercomputer reinforces the race between China and the United States for technological supremacy. American giants such as OpenAI, Anthropic and Google have developed some of the world’s leading AI models, while another US company, Nvidia, has become the leading global supplier of chips for artificial intelligence. China, in turn, has sought to innovate in other ways. Last year, Chinese startup DeepSeek launched an advanced AI model using just a small fraction of specialized chips.
Continues after advertising
To prevent China from catching up with the US, President Donald Trump imposed tariffs and, at times, restricted exports of AI chips. But using standard microprocessors — known as CPUs — instead of GPUs to build an ultrafast supercomputer suggests a possible way around these barriers.
“The U.S. government should impose stricter controls on the export and manufacturing of CPUs for the Chinese market,” said Jimmy Goodrich, a senior fellow at the University of California’s Institute on Global Conflict and Cooperation. “This is a loophole in the current rules.”
Supercomputers — the name given to the largest machines dedicated to science — have been used since the 1960s for tasks such as creating climate models, breaking codes and developing nuclear weapons. In general, they operate with high precision mathematics, expressing numbers with 64 bits of data.
Continues after advertising
Commercial AI systems from companies like Google and OpenAI can be even faster. They use approximations in tasks such as identifying images or choosing the next word in a sentence, using four- and eight-bit numbers, which allows many simpler operations to be carried out at the same time.
“It’s remarkable and impressive what China has done here, but they don’t even come close to these huge AI supercomputers built by American AI labs” and other groups, Goodrich said.
US national laboratories, the main buyers of some of the world’s biggest supercomputers, want to use AI to speed up parts of scientific work. Therefore, they have been adopting more of these less precise calculations, in addition to the traditional 64-bit architecture.
Continues after advertising
Although American companies have historically dominated the top rankings of the largest supercomputers, foreign systems eventually took the lead. A system in Japan, for example, occupied first place on the list between 2020 and 2022.
“There is a lot of talk that America is the only country capable of building these systems,” said Addison Snell, an analyst at Intersect360 Research, a company that follows the sector. “But then you discover that other companies also have this capability.”
Advanced systems from China and Japan have frequently prompted the Department of Energy and other U.S. agencies to push for more funding for supercomputing. In November, the Trump administration launched the Genesis Mission, an initiative that seeks to use supercomputers from US national laboratories, together with private companies, to boost AI and scientific research.
Continues after advertising
GPUs, developed primarily by Nvidia and Advanced Micro Devices, have been a central weapon in the recent supercomputer race. These chips are especially efficient at performing many tasks at once, including vector calculations used in science and matrix multiplication, which is at the heart of many AI applications.
When U.S. officials limited China’s access to GPUs and other powerful chips, as well as restricting the export of some machines used to manufacture the most advanced semiconductors, it led the country to “invest in developing architectures and technology to, in practice, have supercomputers on par with the highest-performing U.S. systems,” Dongarra said.
LineShine does not separate the traditional functions of microprocessors and GPUs, as most high-end systems do. Instead, it incorporates GPU-style tasks through specialized circuits that accelerate matrix and vector calculations. This capacity is embedded in chips that total almost 14 million computing cores — small electronic brains — installed in 90 hardware cabinets.
These chips are an original design based on an instruction set licensed from Arm Holdings, a British company controlled by Japanese conglomerate SoftBank. Arm’s technology is best known for powering smartphones, but it has recently been adapted by Nvidia, Amazon, Qualcomm and other companies for use in data centers.
Dongarra, who wrote a detailed report on the new system, said he was informed during his visit to China that the machine was built without government funding. Therefore, the designers considered that it would be possible to submit the tests to the Top500 ranking.
Shenzhen scientists also sought recognition for the new machine through 14 entries to the Gordon Bell Prize, which promotes the solving of sophisticated scientific problems, Dongarra said. Three systems are finalists for the award, and another three are competing for an award related to climate science.
According to Dongarra’s report, LineShine has already been used in projects such as a sophisticated simulation of the Earth — including atmosphere, ocean, soil and ice — as well as a complex simulation of the human brain.
c.2026 The New York Times Company
