TensorWave deploys North America s largest liquid-cooled AMD GPU training set to attack non-NVIDIA routes

AI cloud-new founding company TensorWave recently announced that it has officially launched a super-large AI training set, composed of 8,192 AMD Instinct MI325X, and is currently the largest fully liquid-cooled AMD GPU system in North America.

TensorWave opens the machine room photo on X, and can see the layout of the magazine orange liquid-cooled pipeline and dense machine cabinets, and emphasizes the identification and maturity of self-developed liquid-cooling technology. Each cabinet contains 8 servers, 8 MI325X is installed in a single cabinet, with a total of 64 GPUs per cabinet, with a power consumption of 100-120 kilowatts, and a complete set of liquid cooling solutions are provided by Supermicro.

8,192 liquid-cooled MI325X GPUs. The largest @AMD GPU training cluster in North America.

Built by TensorWave. Ready for what’s next 🌊 pic.twitter.com/qSGaiPBMM8

— TensorWave (@TensorWaveCloud) July 15, 2025

According to TensorWave Executive Director Derek Horton at the AMD Advancing AI Day event, the focus is on cost and efficiency, with a relatively traditional architecture savings of about 40% and improving AI training efficiency by about 25% to demonstrate the potential of AMD GPUs in large-scale cloud deployments.

MI325X is a flagship AI accelerator released by AMD at the end of last year. It uses a chiplet architecture, built-in 19,456 streaming processors (SP), 256GB HBM3e memory and 6TB/s bandwidth, and has FP8 computing power of 2.6 PFLOPS.

In view of the current market's high dependence on the NVIDIA CUDA architecture, TensorWave hopes to provide another option to create mainstream AI training infrastructure that is not operated by NVIDIA, and break the situation of a single lifelong system.

The company also announced that the next stage will launch a larger-scale MI355X deployment plan and continue to expand its layout in the AMD GPU cloud market, attacking the growth business of future AI computing needs.

TensorWave just deployed the largest AMD GPU training cluster in North America — features 8,192 MI325X AI accelerators tamed by direct liquid-cooling