Distributed inference turns stranded power into AI compute

What happened
Dell Technologies has begun shipping a fully liquid-cooled networking system that integrates compute and storage, designed for large-scale distributed AI inference across multiple sites. The system is delivered as a pretested single rack, reducing installation time from weeks to hours. This infrastructure allows operators to use stranded power sources, such as surplus electricity from various locations, to run AI workloads. The approach unifies cooling and power requirements for AI clusters that can include tens to hundreds of thousands of GPUs.
Key details
- Dell's system is 100% liquid-cooled, including switches, optics, and cabling.
- The solution is shipped as a pretested, single-rack unit, ready for rapid deployment.
- The system supports distributed inference across up to 1,000 small sites with available power.
- Operators can use diverse power sources, including surplus energy from stores or methane capture.
Why it matters
Operators of AI infrastructure can now deploy large-scale AI clusters more quickly and at lower cost by using Dell's unified, liquid-cooled systems. Companies with access to unused or stranded power can convert that energy into AI compute capacity without building traditional data centers. This enables organizations to expand AI capabilities into locations with previously unusable power resources, making it possible to run distributed inference at a much larger scale.