Frore Systems Unveils LiquidJet Cooling for NVIDIA Rubin GPUs at CES 2026
The new coldplate solution targets extreme power densities, supporting up to 1950W TDP for Rubin and 4400W for future Feynman GPUs.
Frore Systems has unveiled LiquidJet, a high-performance coldplate cooling solution designed for the next generation of AI accelerators. Debuting at CES 2026, the technology specifically targets the extreme thermal demands of NVIDIA's Rubin GPUs to prevent performance degradation caused by heat.
The LiquidJet system is engineered to handle massive power loads, offering cooling for up to 1950W TDP on NVIDIA Rubin GPUs, which feature two reticle-sized dies and 8x HBM. According to technical specifications, the solution can maintain a chip junction temperature (TJmax) of 80.5°C when the liquid inlet temperature is 40°C. To achieve these results, Frore utilizes a multistage cooling architecture featuring a 3D short-loop jet-channel structure paired with PTM 7950 thermal interface material (TIM).
The Thermal Bottleneck
As AI GPUs evolve, power densities are skyrocketing, with thermal design power (TDP) now frequently exceeding 1000W. Traditional liquid cooling methods are increasingly unable to keep junction temperatures below the critical throttling threshold. This has led hyperscalers to explore "delidding"—the removal of the integrated heat spreader—to place coldplates directly onto the silicon. While once a niche practice for extreme overclockers, direct-to-die cooling is becoming a production necessity for massive AI clusters to maintain stability and speed.
Scaling for the Future
Thermal management has become the primary bottleneck for scaling AI compute. By reducing GPU junction temperatures by up to 7.7°C, Frore aims to maximize the return on investment for expensive NVIDIA hardware. Efficient thermals not only prevent throttling but also improve the overall Power Usage Effectiveness (PUE) and Total Cost of Ownership (TCO) for data center operators who must manage the immense heat output of thousands of concurrent accelerators.
Next-Generation Targets
Frore is positioning LiquidJet as a future-proof investment for the AI roadmap. The company stated that the technology is designed to scale up to 4400W of cooling capacity to support the upcoming NVIDIA Feynman GPUs. While Tom's Hardware reported company claims that this thermal headroom could boost performance by 15%, that specific figure remains unverified by independent technical sources. The industry will now be watching to see if these lab-demonstrated TDP targets translate to stable, long-term deployments in production AI environments.