Chinese start-up tackles fusion energy software bottleneck with help of AI
For decades, fusion energy has sat at the edge of humanity’s clean-energy imagination – a promise of virtually limitless power that always seems just out of reach. But for Xie Huasheng, a fusion theorist and plasma simulation scientist, the industry finally has a tangible way to shorten its costly t
By Wency Chen
For decades, fusion energy has sat at the edge of humanity’s clean-energy imagination – a promise of virtually limitless power that always seems just out of reach. But for Xie Huasheng, a fusion theorist and plasma simulation scientist, the industry finally has a tangible way to shorten its costly trial-and-error cycle: better software.
“Fusion simulation software has long faced an ‘impossible triangle’,” Xie said. Existing tools, he argued, tend to be either accurate but computationally expensive, fast but unreliable, or simple in principle but too crude to allow accurate extrapolation and guide next-generation reactor design.
“We are now at a turning point,” Xie said, noting that the performance of more than a dozen physics design and analysis models had improved sharply, driven by refined mathematical structures and advances in artificial intelligence that improved research efficiency.
In April, Xie founded VeloAlpha, a Beijing-based start-up building “FusionAlpha” – a simulator designed to help developers test reactor designs on computers before committing to expensive physical experiments.
He likened it to electronic design automation (EDA) software in the semiconductor industry, where chipmakers test and optimise designs long before they go to the wafer foundry.
Fusion is the reaction that powers the sun, releasing massive amounts of energy when nuclei of light atoms are forced to collide and merge. To recreate this cosmic process on Earth, scientists have to heat fuel until it becomes an extremely hot, electrically charged gas called plasma, and hold it stable long enough for the reaction to keep going.
The leading machine for this task is a tokamak – essentially a giant, doughnut-shaped bottle designed to trap plasma using powerful magnetic fields. But it is not the only design: scientists are also experimenting with twisted magnetic tracks known as stellarators, straight magnetic tubes, and high-powered lasers.
Regardless of the approach, the industry faces immense challenges: proving it can sustain reactions, managing extreme heat and radiation, securing fuel supplies and generating electricity cheaply enough to compete with existing sources. Historically, attempting to clear these hurdles has required repeated, financially draining physical experiments.
Software has not always been able to help. For years, scientists relied on massive, hyper-detailed physics programmes to simulate these reactions. While highly accurate, these programmes were extremely slow and costly to run.
On the flip side, newer tools powered by AI – like Google DeepMind’s simulation software – compute much faster but can be less reliable, while simplified models offer only rough guidance.
Xie, a Zhejiang University-trained plasma physicist who conducted postdoctoral research at Peking University, is determined to tackle this software bottleneck.
He previously served as chief scientist for fusion theory and simulation at the ENN Energy Research Institute, one of China’s earliest private fusion pioneers, spending much of his career developing mathematical and software tools for modelling fusion plasmas.
Xie said FusionAlpha broke traditional constraints by using new mathematical methods that respect the laws of physics but strip away the digital lag. He claimed that some of his software’s core features can run anywhere from 100 to 10,000 times faster than the industry’s current state-of-the-art fusion codes.
Even at that breakneck speed, Xie said the software kept benchmark calculation errors under 5 per cent, but he added that the numbers still needed to be verified by outside experts.
VeloAlpha plans to release its first full-process simulator, starting with tokamak settings, next year. Several core modules have already been open-sourced, a strategic move aimed at building a user base and establishing trust within the tight-knit fusion community.
Over time, Xie envisioned FusionAlpha becoming the standard design platform for fusion start-ups, state-backed laboratories and hardware suppliers alike.
Investors are buying into the vision. VeloAlpha recently closed a seed funding round of tens of millions of yuan, led by Loongson Venture Capital, with participation from Zhejiang University-linked venture firm Oufang Angel and a state-backed fund managed by Legend Capital. Its next funding round is close to completion, Xie said.
The start-up’s emergence coincides with a shift in China’s fusion sector, which is transitioning from a largely state-led research effort into a broader commercial ecosystem.
Beijing has officially designated hydrogen and nuclear fusion as “future industries” targeted for high-stakes growth under its 15th Five-Year Plan, placing it alongside strategic sectors like quantum technology, biomanufacturing, brain-computer interfaces, embodied AI and 6G.
This strong policy signal has triggered a wave of venture capital. Commercial fusion start-ups like Startorus Fusion, Energy Singularity, Nova Fusion and Dongsheng Fusion have raised increasingly large rounds, while suppliers of magnets, power systems and advanced materials are rapidly positioning themselves around the ecosystem.
Though commercial fusion remains highly uncertain, its strategic allure is undeniable.
“The next five to 10 years will be critical for fusion,” Xie said. “The pace of iteration is accelerating. The future of fusion won’t just be built in steel and concrete, but first in high-precision digital codes.”