VeloAlpha, a Beijing-based start-up founded in April 2026, is building an AI-powered fusion reactor simulation platform called FusionAlpha, aiming to cut the costly trial-and-error cycle that has long slowed progress toward commercially viable fusion energy. The company was founded by Xie Huasheng, a fusion theorist and plasma simulation scientist, who argues that smarter software — not just bigger machines — is the missing lever in the global race to harness fusion power.

The 'Impossible Triangle' of Fusion Software

According to Xie Huasheng, fusion simulation tools have historically been trapped in what he calls an 'impossible triangle': existing platforms tend to be either accurate but computationally expensive, fast but unreliable, or conceptually simple but too crude to guide next-generation reactor design. No single tool has managed to satisfy all three demands simultaneously.

Xie said the performance of more than a dozen physics design and analysis models has improved sharply, driven by refined mathematical structures and advances in artificial intelligence that have improved research efficiency. 'We are now at a turning point,' he said.

The EDA Analogy: Lessons from Semiconductors

Xie likened FusionAlpha to electronic design automation (EDA) software in the semiconductor industry, where chipmakers simulate and optimise designs on computers long before committing to expensive wafer fabrication. The parallel is deliberate: EDA tools became a foundational layer of the chip industry, and VeloAlpha is betting fusion will follow the same trajectory.

Fusion energy works by forcing nuclei of light atoms to collide and merge — the same reaction that powers the sun — releasing massive amounts of energy. To replicate this on Earth, scientists must heat fuel to extreme temperatures until it becomes plasma, an electrically charged gas, and hold it stable long enough to sustain the reaction. Simulating this process accurately and cheaply has remained one of the field's hardest unsolved engineering problems.

Backers and Institutional Links

The start-up has attracted backing from investors including Loongson Venture Capital, Legend Capital, and Oufang Angel, according to reports. VeloAlpha also has institutional ties to Peking University and Zhejiang University, and has drawn interest from energy-sector players including ENN Energy Research Institute. The involvement of Google DeepMind in the broader fusion AI ecosystem has also been noted by industry observers as a signal of how seriously frontier AI labs are treating the sector.

Why It Matters

Fusion has been described as perpetually '30 years away' for most of its modern history, but a convergence of private capital, AI tooling, and advances in plasma physics is compressing timelines in ways that were not plausible a decade ago. Simulation software that meaningfully reduces physical experimentation costs could shift the economics of fusion development, making smaller, better-capitalised teams competitive with state-backed programmes.

As the global fusion industry moves from proof-of-concept to engineering-scale challenges, the companies that own the simulation stack — much like EDA vendors in semiconductors — could capture outsized value regardless of which reactor design ultimately wins.

What's Next

The immediate test for VeloAlpha is whether FusionAlpha can demonstrate measurable accuracy gains over incumbent tools while remaining computationally tractable for commercial reactor developers. With seed funding secured and a founding team rooted in academic plasma physics, the company's next milestones will likely centre on pilot partnerships with reactor developers and further validation of its AI-enhanced models. The broader fusion software market is nascent but accelerating — and VeloAlpha is moving early.