In response to the explosive growth of AI and the escalating thermal demands of next-generation GPUs, we are excited to announce LiquidJet™, a revolutionary direct-to-chip liquid cooling solution designed to meet the needs of AI Data Centers. Built with Frore's unique semiconductor manufacturing process adapted to metal wafers, LiquidJet unlocks higher performance from the world's most powerful GPUs—starting with NVIDIA Blackwell Ultra—delivering significant reductions in data center total cost of ownership (TCO).
LiquidJet's breakthrough 3D short-loop jet channel microstructures deliver game-changing thermal performance, enabling data centers to push GPUs harder, achieving more AI tokens per second, and improving power usage effectiveness (PUE). Removing heat efficiently unleashes the AI performance needed for training, inference, and the full spectrum of data center workloads powering the AI revolution.
An analysis by McKinsey & Company suggests “that demand for AI-ready data center capacity will rise at an average rate of 33 percent a year between 2023 and 2030 (reflecting a trend that is already underway.)” This expansion is transforming data center design, resulting in increased power consumption and new cooling strategies.
LiquidJet is already demonstrating delivering game-changing results with the NVIDIA 1,400W Blackwell Ultra
LiquidJet outperforms traditional coldplates on every major cooling metric:
- 2x higher hotspot power density (600 W/cm² @ 40ºC inlet temp)
- 50% higher KW/lpm (cooling efficiency)
- 4x lower pressure drop (requiring less pumping power)
Future-proofed for what's next, LiquidJet is engineered to scale with next-generation SoCs like NVIDIA Rubin, Rubin Ultra, Feynman (>4,000W) and beyond, as well as custom hyperscaler ASIC designs. As AI workloads explode and chip power continues to climb, Frore's LiquidJet architecture is built to keep up.
Heat removal, which limits performance, is now one of the biggest challenges facing AI data centers. LiquidJet's ability to dissipate extreme heat loads, unleashing GPU performance, is essential for maximizing AI infrastructure ROI.
A New Approach to Coldplate Manufacturing
Legacy coldplates rely on outdated skiving and 2D microchannel manufacturing, limiting their ability to adapt to new chip layouts and the increasingly non-uniform, extreme power density requirements of modern AI GPUs.
Frore Systems takes a radically different approach, adapting semiconductor manufacturing to metal wafers, and fabricating 3D short-loop jet channel microstructures that map precisely to the power distribution of modern GPUs. This enables:
- Highly customized designs to exactly match any SoC power map
- Far superior cooling performance through optimized 3D fluid dynamics
- Scalable, cost-effective manufacturing using proven semiconductor processes
- Easy drop-in upgrades compatible with existing liquid cooling infrastructure
- Rapid design iterations to match the accelerating pace of AI chip development
The result is LiquidJet—a coldplate that evolves as fast as the chips it cools.
As Seshu Madhavapeddy, our Founder and CEO stated, "LiquidJet's unique 3D architecture, with customized 3D short-loop jet channel microstructures, sets a new bar for coldplate thermal performance. Just as AirJet redefined active cooling for consumer and edge devices, LiquidJet transforms coldplates into a future-ready platform for the AI Factory."
LiquidJet enables AI data centers to achieve:
- Cooler GPUs – Lower junction temperatures for better reliability and longer lifespan
- More AI tokens/second – Eliminate thermal throttling and unlock full chip performance
- Lower Total Cost of Ownership (TCO) – Reduce cooling infrastructure and energy costs
- Improved Power Usage Effectiveness (PUE) – Increased efficiency gains across the facility
- Higher rack densities – Pack more AI compute into existing footprints
- Future-ready scalability – One cooling platform that grows with your AI roadmap
LiquidJet makes maximizing AI performance a breeze, enabling data centers to extract more value from every GPU, every watt, and every square foot of data center space.
Precision-Mapped Cooling: The LiquidJet Advantage
LiquidJet's 3D short-loop jet channel architecture fundamentally reimagines liquid cooling:
- Precision-Mapped Cooling: Using semiconductor manufacturing processes adapted to metal wafers, LiquidJet creates 3D short-loop jet channels positioned exactly where GPU hotspots occur. Coolant flows through optimized 3D paths, targeting high power density regions with surgical precision.
- 3D Short-Loop Jet Channels: Unlike traditional parallel microchannels where coolant temperature rises along the channel length, LiquidJet uses 3D short-loop jet channels that impinge directly on hotspot regions before recirculating. This maintains consistently cold coolant at critical locations, maximizing heat removal.
- Adaptive Design Platform: The same semiconductor fabrication process that creates LiquidJet for today's Blackwell Ultra can be rapidly adapted for tomorrow's Rubin, Feynman, or custom ASIC designs. As GPU architectures evolve, LiquidJet evolves with them—without costly retooling.
- Drop-in Compatibility: Despite its revolutionary internal architecture, LiquidJet maintains compatibility with standard CDU connections, manifolds, and mounting systems. Data centers can upgrade to breakthrough performance without infrastructure overhaul.
Sustained Performance and Scalability
LiquidJet is engineered for the extreme demands of AI data centers, where GPUs run at maximum capacity 24/7/365. The 3D short-loop jet channel architecture ensures consistent cooling performance even as GPU power continues to scale beyond 4,000 watts per chip.
Future-proof supporting NVIDIA Feynman SoC & beyond
LiquidJet's semiconductor manufacturing process enables rapid scaling and customization for any chip layout, power map, or thermal requirement. Whether cooling NVIDIA's latest GPUs, custom hyperscaler ASICs, or future chips not yet designed, LiquidJet's platform adapts quickly and cost-effectively.
All LiquidJet needs to enable exceptional GPU performance is integration with standard liquid cooling infrastructure. The coldplate does the rest, delivering superior thermal performance that unleashes AI compute capability.
There are a massive number of AI data centers searching for the best way to maximize GPU performance
AI is transforming data centers into high-throughput AI Factories, running massive training workloads and serving billions of inference requests. This requires running increasingly powerful GPUs at maximum performance in increasingly dense configurations—which inevitably means extreme thermal challenges. LiquidJet unleashes the enhanced performance needed across AI applications including:
- Large language model training
- AI inference at scale
- Computer vision and image generation
- Autonomous vehicle development
- Scientific computing and simulation
- Custom AI model development
- Next-generation AI research
- And more…...
In today's AI factory data centers, what often determines performance is the capability of the thermal solution, not just the sophistication of the GPU. With LiquidJet, AI factories can now deliver on the full promise of cutting-edge AI accelerator technology—today and well into the future.
The AI revolution demands revolutionary cooling. LiquidJet delivers.
