Frore LiquidJet Diamond Makes AI Server Cooling Part of the Performance Stack

As of October 12, 2026, Frore Systems’ LiquidJet Diamond is no longer just a behind-the-scenes thermal component announcement. The company introduced the coldplate on October 7, 2026, and is showing it during OCP Global Summit week in San Jose, where AI infrastructure cooling is becoming as important as compute density. LiquidJet Diamond is aimed at AI factory hardware, not consumer PCs, and its job is straightforward: move heat away from the most intense GPU hotspots before those hotspots limit sustained performance. Frore says the design can improve AI Tokens/Watt by 30% to 35% compared with traditional skived coldplates, but those figures should be treated as vendor claims rather than independent production benchmarks. (frore.com)

LiquidJet Diamond is a direct-to-chip liquid-cooling coldplate with an integrated diamond spreader. Frore describes it as a 3D ultra short-loop, multi-stage coldplate that can be customized to the power map of each GPU, directing coolant toward hotspot areas before cooling the rest of the GPU die, HBM, and other chiplets. The diamond is not mixed into the coolant; it is embedded into the bottom surface of the coldplate and positioned directly above hotspots to spread heat into a larger area before the liquid-cooling structure takes over. Frore’s listed construction uses copper and diamond, and the company says dimensions are customized per target GPU. For an NVIDIA GB300 example, Frore lists measurements of 104.6 mm long, 62 mm wide, and 20.4 mm high. (frore.com)

View NVIDIA DGX Spark Personal AI Desktop Supercomputer on partner website

The standout number is hotspot cooling density of up to 770 W/cm², specified for a 5 mm x 5 mm hotspot with a 25 °C inlet temperature, 95 °C TJmax, PTM7950 thermal interface material, and an unlidded package. Frore also claims LiquidJet Diamond can deliver over 150% higher maximum hotspot power density, 75% higher heat transfer efficiency, and a 16 °C to 22 °C lower GPU temperature compared with traditional skived coldplates. Those are very specific comparison claims, so the context matters: real-world results will depend on the accelerator package, coolant loop, flow rate, inlet temperature, thermal interface, and rack design. (frore.com)

See ASUS GeForce RTX 5090 price

The timing is important because next-generation AI accelerators are pushing thermal design into less comfortable territory. Frore says it is demonstrating LiquidJet cooling for a 2300W NVIDIA Rubin setup during the OCP event, while LiquidJet Diamond demonstrations focus on the 770 W/cm² extreme hotspot target. Tom’s Hardware and eeNews Europe both framed the product around high-power AI processors, with the Rubin-class angle making it clear this is a data center cooling story rather than a flashy waterblock for gaming PCs. In these systems, cooling is not only about preventing shutdowns; it can influence leakage, throttling behavior, and how much useful AI throughput a rack can deliver under a fixed power budget. (frore.com)

Buy Lenovo Legion Pro 7i laptop with Core Ultra 9 and NVIDIA RTX 5090 24GB here

LiquidJet Diamond is interesting because it treats cooling as part of the performance stack. AI infrastructure planning has often focused on GPU count, networking, and power delivery, but dense accelerators make thermal engineering a first-order design problem. Frore’s approach suggests a future where coldplates are mapped more closely to silicon layouts, with hotspot-specific features becoming normal in high-end AI servers. The product still needs broader validation in commercial deployments, especially around cost, reliability, manufacturability, and long-term serviceability. Even so, the direction is clear: as AI racks become more power-dense, the best cooling systems may help determine not just whether hardware runs safely, but how efficiently it converts watts into tokens.

See SANDISK Optimus GX PRO 8100 PCIe 5 SSD price