A $3,000 DIY fusion reactor built with Claude
A Canadian maker named Hudzah just built a tabletop nuclear fusor capable of detecting neutrons for around $3,000, guided largely by Anthropic's Claude and OpenAI's o1 pro. Claude Projects digested forum archives to debug the build, while o1 pro handled intricate electrical wiring. Why it matters: Frontier reasoning models are actively lowering the barrier for complex, high-risk physical engineering at home.

Building a working nuclear reactor on your kitchen counter used to require a PhD, a massive budget, and years of scouring dusty textbooks. Thanks to generative AI, a Canadian maker just pulled it off for roughly $3,000 in a single weekend.
A maker known online as Hudzah recently livestreamed his homebuilt fusion reactor on X. His goal was to construct a functional fusor that could actually fuse atoms and emit measurable neutrons, moving far beyond a simple plasma demo he built months ago.
Toy cars, heavy water, and an AI co-pilot
To pull this off, Hudzah scavenged parts on eBay, navigated Canada Post strikes, and found clever workarounds for expensive scientific gear.
Instead of ordering industrial deuterium gas, he bought a $32 toy hydrocar equipped with a tiny proton exchange membrane electrolysis cell. He then fed it $80 worth of heavy water, producing 56 liters of deuterium gas that he stored in medical syringes.
The final setup featured a 30kV/10mA electrostatic power supply, a specialized bubble counter to track neutron emissions, and a vacuum chamber capable of reaching 3 millitorr, which is roughly 253,000 times more vacuum than normal atmospheric pressure.
Assembly was where the AI came in. Hudzah created a massive project in Anthropic's Claude, packing it with forum documentation, Granola call transcripts, and email threads to guide debugging and safety protocols. When it came to confusing electrical wiring and physical assembly, he turned to OpenAI's o1 pro to clear up the bottlenecks. Armed with AI answers, he speedran the entire build in less than 48 hours.
Why it matters
We think this project is a startling preview of how frontier models are dismantling the barrier between digital knowledge and real-world execution.
For independent makers and hardware hackers, this is a massive unlock. Translating dense academic papers and messy forum threads into working, high-voltage hardware usually takes years of trial and error. By pairing Claude's context window with o1 pro's deep reasoning, a lone hobbyist essentially had an on-demand team of lab technicians working the bench with him.
However, it also exposes a glaring safety tightrope that AI labs will soon have to confront. A 30kV electrical supply and tabletop nuclear fusion are inherently lethal if mishandled in a home kitchen. While the models handled safety queries here, providing step-by-step assembly guidance for hazardous physical systems will inevitably trigger scrutiny from regulators and force model makers to rethink their guardrails around physical-world experimentation.
Sources (2)
- @hud_zah(opens in a new tab)Originaltwitter.com
- tweaktown.com(opens in a new tab)Reporttweaktown.com