Liqid Inc. is introducing a memory pooling technology designed to optimize AI inference infrastructure. By decoupling memory and GPUs from fixed server configurations, the company allows data centers to dynamically allocate resources based on the specific requirements of individual AI models.

According to co-founder and CTO Sumit Puri, the approach addresses hardware supply chain constraints by ensuring that applications receive only the necessary amount of memory. This efficiency is measured by Liqid through metrics such as tokens per second, tokens per watt, and tokens per dollar.

At the SC26 conference, the company announced the launch of Liqid as a Service. This new offering provides a fully managed, single-tenant environment for enterprises. The company is also shifting toward preconfigured appliances to support businesses that prioritize model deployment over building infrastructure from scratch.