Bentaus, an energy‑first NeoCloud, teamed with DriveNets to deploy the DriveNets AI Fabric on a cluster of AMD Instinct MI325X GPUs. The joint solution uses Bentaus’ Ziani orchestration software to adjust GPU power in real time. According to the companies, the setup reduces energy per token by 30% and keeps 97% of token throughput, meaning only a 3% loss in speed. The architecture also claims about 5% higher overall throughput and a 12% faster time to the first token compared with publicly available results.
Why it matters
Enterprises can run more AI work for less electricity, easing grid strain in regions with data‑center power limits.
