Frore Systems has introduced the LiquidJet Diamond water block for server liquid cooling systems, designed specifically for use in artificial intelligence data centers. The new product is unique due to the presence of a built-in diamond heat spreader.
Image source: froresystems.com
Growing demand for AI infrastructure coupled with energy shortages is driving hyperscalers to look for new ways to improve data center efficiency. According to the manufacturer, the LiquidJet Diamond water block will allow systems to generate 35% more tokens per watt of energy consumed, and increase revenue by the same amount compared to traditional microchannel water blocks.
New LiquidJet Diamond is built on the basis of the previously presented Frore Systems LiquidJet water block – it uses a multi-stage 3D system with an ultra-short circulation circuit, which can be adapted to the best heat removal from the hottest areas of each GPU. Such precision becomes critical as the complexity and power consumption of GPUs grows to meet the needs of AI. LiquidJet helps reduce GPU die temperatures by 12°C, which translates to a 25% increase in efficiency on a token-per-watt basis.
The LiquidJet Diamond model adds another heat spreader to the bottom of the system with built-in diamond plates that target hot spots. At these points, the temperature drops by another 10 °C, and the final efficiency figure increases to 35%.
“Combining an innovative ultra-short loop 3D multi-stage design with an integrated diamond heat spreader, LiquidJet Diamond is capable of cooling hot spots with a heat flux of up to 770 W/cm². This unlocks the potential of high-performance GPUs for AI and provides even higher token yield per watt. If each AI data center could produce 35% more AI tokens with the same resources by implementing LiquidJet Diamond, hyperscalers would achieve their goals faster and with fewer data centers.”,” said Frore Systems founder and CEO Seshu Madhavapeddy.
Compared to traditional microchannel water blocks, Frore LiquidJet Diamond cools hot spots with over 150% higher heat density; heat transfer efficiency is 75% higher; GPU temperature is reduced by 16 °C to 22 °C; token yield per watt increases by 30–35%. Installing new cooling systems does not require a complete rework of the system; Finally, Frore Systems offers scalability to meet data center needs in both size and production volume.
The traditional method of producing water blocks with fins milled inside to create microchannels for coolant no longer provides the high enough heat dissipation performance required by AI GPUs. 3D metal printing (ECAM) technology provides design flexibility, but cannot meet hyperscalers’ demands for quality, reliability, production volume or lead time.
Frore Systems says LiquidJet offers semiconductor manufacturing-level precision with mass production capabilities—these cooling systems meet both the performance and scalability needs of data center LSS applications, Frore Systems said.
The manufacturer is ready to adapt products to specific accelerator models: for example, Nvidia Blackwell Ultra crystals have an area of 1488 mm2, and Cerebras Wafer-Scale Engine is already 46,225 mm2 or 30 times larger.
Frore Systems promised to present its latest LiquidJet and LiquidJet Diamond systems at the OCP Global Summit, working in conjunction with Nvidia Rubin chips with a power of 2300 W – they will provide a heat flux of 770 W / cm²; the integrated LiquidJet Nexus system will demonstrate the cooling of ½ U form factor computing modules. The event will take place from October 12 to October 16, 2026.
If you notice an error, select it with the mouse and press CTRL+ENTER.
