Startup activates nuclear microreactor live on stage to power an Nvidia RTX Spark desktop PC — firm working with Nvidia to build a 30MW closed loop AI factory that doesn't use local water

Tom's Hardware reports on a stunning demonstration at an industry event where a startup activated a working nuclear microreactor live on stage to power an NVIDIA RTX Spark desktop PC — a showcase designed to make a provocative point: that the next generation of AI computing infrastructure, from on-device inference to data center-scale training, will require a fundamental rethinking of power delivery. The demonstration featured a compact microreactor, small enough to fit on a stage, directly powering an RTX Spark N1X-based desktop system running local AI inference workloads — proving that integrated Arm-based superchips paired with modular nuclear power could redefine where and how AI compute happens. The startup, whose identity was revealed during the presentation, is also working with NVIDIA on plans to build a 30-megawatt closed-loop AI factory that eliminates local water consumption entirely, using advanced cooling and heat-reuse systems to address two of the most significant environmental constraints facing the rapidly expanding AI data center industry.

Tom's Hardware details the technical specifics of the microreactor demonstration, noting that the unit generates enough electricity to power not just the RTX Spark desktop but multiple systems simultaneously, while operating within a footprint comparable to a shipping container. The closed-loop design recirculates cooling fluid rather than drawing from local water sources, addressing a growing concern as AI inference clusters proliferate in water-scarce regions. The 30MW closed-loop factory planned in partnership with NVIDIA would be capable of hosting thousands of RTX Spark nodes or conventional GPU servers, all running without consuming local water — a design that could prove decisive in regions where data center water usage has become a flashpoint for community opposition. The article notes that this approach mirrors broader industry trends toward co-locating compute with power generation, as the latency requirements of real-time AI inference and agentic workloads make centralized cloud architectures increasingly impractical for many use cases.

The demonstration carries significant implications for RTX Spark's positioning as more than just a consumer PC platform. By showing an RTX Spark desktop running on decentralized nuclear power, NVIDIA is signaling that the platform's efficiency advantages — including the N1X chip's 45-80W power envelope and unified memory architecture — make it suitable for edge AI deployments in environments where traditional data center infrastructure is unavailable or impractical. The Tom's Hardware report frames this as part of NVIDIA's broader strategy to make AI inference ubiquitous, from desktop PCs to edge servers to compact data centers, all powered by the same software stack and architectural principles. For Canadian readers, the implications are particularly relevant: Canada's nuclear expertise and abundant uranium resources, combined with the country's growing AI sector and concerns about data center water usage in regions like southern Ontario and Quebec, position RTX Spark as a potentially transformative platform for cost-effective, sustainable local AI computing.