Liqid Inc., a Jeffco-based technology company, is heading to the world’s largest AI conference next week to show off a new way to make data centers faster and more efficient. At NVIDIA GTC 2026, the company will demonstrate its Liqid Matrix software, which acts like a digital conductor for computer hardware. Instead of high-powered chips being locked inside a single computer, Liqid’s technology allows them to be shared and moved instantly between different machines as needed. This pooling of resources helps companies run powerful AI programs more cheaply and with much less wasted energy.
The event, widely regarded as the world’s premier conference for AI and accelerated computing, will take place from March 16-19 at the San Jose Convention Center. Liqid will be stationed at Booth #121, offering live demonstrations of the industry’s only multi-fabric platform capable of dynamically scaling and sharing resources across standard servers in real-time.
As AI workloads become increasingly resource-intensive, data centers are seeking ways to maximize hardware efficiency. Liqid’s composable solutions address this by allowing organizations to pool expensive resources like GPUs and CXL memory, then assign them to specific servers on demand.
According to the company, this approach delivers significant performance and economic gains, including:
- 2x more tokens per watt compared to traditional static configurations.
- 50% higher tokens per dollar, significantly lowering the total cost of ownership for AI infrastructure.
At the heart of the demonstration is the Liqid Matrix fabric manager. This software allows administrators to orchestrate pooled resources and integrates natively with popular industry tools like Kubernetes, NVIDIA NIM, Nutanix NKP, and Slurm.
The showcase will highlight two primary solutions:
- Composable GPU Solutions: This technology enables the connection of up to 30 GPUs to a single server. Once a task—such as high-demand AI inference—is completed, the GPUs can be instantly reconfigured and redeployed to other workloads.
- Composable CXL Memory: Designed for memory-intensive applications, this solution provides the flexibility needed for next-generation AI inference, in-memory databases, and high-performance computing (HPC) workloads.
Liqid experts will be on-site to show how these technologies optimize a variety of modern enterprise needs, including KV cache, Virtual Desktop Infrastructure (VDI), and virtualization. By maximizing infrastructure utilization, Liqid aims to help organizations move past the limitations of traditional, siloed hardware setups.
Attendees of NVIDIA GTC 2026 can visit the Liqid booth to test drive the Liqid Matrix software and see firsthand how composability is reshaping the future of the enterprise data center.
Source: Press Release
