AI accelerators are getting hot enough that the coldplate itself is becoming part of the performance strategy. Frore Systems is using the 2026 OCP Global Summit to demonstrate LiquidJet Diamond, a direct-to-chip coldplate designed for next-generation GPUs with extreme local heat density.
Frore says LiquidJet Diamond combines its 3D ultra-short-loop, multi-stage liquid channels with an integrated synthetic-diamond heat spreader positioned over GPU hotspots. At OCP, the company is demonstrating the design against a 2,300-watt NVIDIA Rubin-class thermal target and hotspot densities reaching 770 W/cm².
Water-cooling enthusiasts are already debating whether adding diamond materially improves real systems, with several commenters correctly focusing on interface resistance and the need for independent testing.
The thermal logic is straightforward. Diamond can conduct heat far more effectively than copper, allowing heat from a small hotspot to spread across a larger area before it reaches the coolant channels. Frore claims the Diamond version can keep GPUs 16–22°C cooler and deliver 30–35% more AI tokens per watt than traditional skived coldplates. Those are company-reported figures, not independent benchmarks, so they should be treated as performance claims rather than established fleet economics.
This CES 2026 walkthrough shows Frore’s original LiquidJet architecture, including the short-loop jet-channel approach used to concentrate coolant around GPU hotspots.
The bigger engineering issue is that accelerator power is rising faster than conventional rack cooling can comfortably absorb. BitcoinVersus.Tech recently covered OCP’s expectation that AI racks will move beyond 1 MW. At that density, reducing junction temperature is not just about preventing failure. Cooler silicon can hold higher clocks with less leakage and thermal throttling, potentially increasing useful compute from the same electrical service.
Liquid cooling is already spreading deeper through the rack. KIOXIA has brought liquid cooling to enterprise SSDs, while Delta is showing multi-megawatt cooling alongside 800 VDC power. LiquidJet Diamond pushes that trend down to the chip package itself.
Frore’s broader CES 2026 booth tour connects LiquidJet’s direct-to-chip approach with the company’s other thermal technologies.
Tom’s Hardware notes that Frore has not disclosed every detail of how the diamond layer is integrated or manufactured. That matters because thermal interfaces, bonding quality, coolant flow and pressure drop can determine whether excellent material properties translate into better rack-level performance.
The hardware is still worth watching because the problem is real. As accelerator modules move toward multi-kilowatt power levels, data-center operators cannot solve every thermal bottleneck simply by increasing water flow. Better heat spreading and more precise coolant delivery may become as important as bigger pumps and larger chillers.
LiquidJet Diamond is therefore best viewed as a thermal-engineering bet: use a better conductor exactly where the GPU is hottest, then use microchannels to remove that heat quickly. Whether the claimed 30–35% tokens-per-watt gain survives independent testing will determine how meaningful the diamond layer becomes outside the OCP demo floor.
BitcoinVersus.Tech Editor’s Note:
We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support to help further secure the integrity of our research initiatives, please donate using the QR code provided
For best QR-code scanning results, please scan this donation code from a desktop or larger screen. Mobile screens may prevent camera recognition.
Leave a Reply