# Frore debuts LiquidJet Nexus coldplate for AI accelerators > Source: > Published: 2026-06-04 11:53:12.010905+00:00 # Frore debuts LiquidJet Nexus coldplate for AI accelerators Frore Systems unveiled the **LiquidJet Nexus**, a monolithic water block designed to cool two Blackwell GPUs and a Grace CPU, Tom's Hardware reports. Per Tom's Hardware, tests run by a major ODM showed the LiquidJet Nexus cut GPU temperatures by about **6C** versus default cooling solutions and produced a **10%** increase in token generation in those tests. The product uses semiconductor-manufacturing techniques such as **etching and bonding** to create coldplates matched to thermal maps, rather than traditional milled coldplates, Tom's Hardware says. Frore is demonstrating a version for Nvidia's **Grace Blackwell** and Tom's Hardware reports a Vera Rubin-compatible version is coming. ### What happened Tom's Hardware reports that Frore Systems is showing the **LiquidJet Nexus** at an event in Taipei, Taiwan. The LiquidJet Nexus is a monolithic water block designed to cool two Blackwell GPUs and a Grace CPU inside a server tray, Tom's Hardware says. According to Tom's Hardware, tests conducted by a major ODM found the LiquidJet Nexus reduced GPU temperatures by roughly **6C** compared with the default cooling solution and delivered a **10%** increase in token generation in those tests. Tom's Hardware also reports Frore demonstrated a version for Nvidia's **Grace Blackwell** and that a version compatible with **Vera Rubin** is forthcoming. ### Technical details Tom's Hardware writes that Frore's LiquidJet coldplates are manufactured using semiconductor fabrication techniques, including **etching and bonding**, and are architected using actual thermal maps of the processors they target. The article contrasts these coldplates with traditional milled coldplates, arguing the semiconductor-tooling approach removes heat more precisely from processor hotspots, which Tom's Hardware links to the reported temperature and token-generation gains. ### Industry context Editorial analysis: Companies that apply semiconductor manufacturing methods to thermal hardware aim to improve precision at scale, which can yield smaller thermal resistance and tighter per-die control. For operators running dense AI racks, even single-digit Celsius reductions in hotspot temperature can translate into sustained performance gains or improved power efficiency, according to industry performance math common in data-center cooling design. ### What to watch Editorial analysis: Observers should look for independent benchmark reports and replication of the ODM test results, compatibility details for different accelerator SKUs, and coolant-loop integration requirements for rack-level deployments. Also watch manufacturability and supply timing, since Tom's Hardware notes Frore uses semiconductor fabs and bonding steps that differ from conventional coldplate production. ### Limitations Tom's Hardware attributes the performance numbers to tests by a major ODM; the article does not provide raw datasets or independent third-party verification. Tom's Hardware is the sole source for the claims in this report. ## Scoring Rationale This is a notable hardware innovation for AI infrastructure that could improve accelerator cooling and performance. The story hinges on ODM test results reported by a single outlet; independent verification will determine broader operational impact. Practice interview problems based on real data 1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)