RA'ANANA – November 12, 2026 -- Bentaus, an energy-first NeoCloud, cut GPU power by 30% while retaining 97% of token throughput on an AI cluster built with DriveNets networking, according to a joint announcement by the two companies.
The cluster runs AMD Instinct MI325X GPUs on DriveNets AI Fabric as the back-end network. It is managed by Bentaus Ziani Systems, which uses the GPU-FLX and CPU-FLX platforms to adjust compute power in real time. The companies describe the result as a grid-responsive, vendor-agnostic infrastructure built on open standards, aimed at maximizing tokens per watt and lowering cost per token.
Ziani software curtails power in real time without halting AI workloads
Ziani lets data centers reduce GPU and CPU power draw dynamically and return power to the grid while AI workloads keep running. Bentaus CEO and Founder Bob Davidoff said the aim is "more compute, delivered more efficiently, on power the grid can sustain."
DriveNets fabric targets GPU utilization from switch to kernel
DriveNets' Ethernet-based fabric optimizes the full networking stack, from switch and NIC to RCCL and kernel. It is tuned for AMD environments and based on a reference architecture engineered jointly with AMD. In published benchmarks, the architecture delivered roughly 5% higher throughput and 12% better time to first token than publicly available results, according to the companies.
State moratoriums on data centers raise the stakes for power-flexible compute
More than a dozen U.S. states have introduced data center moratoriums or paused legislation in 2026, citing grid strain and ratepayer costs. New York has enacted a statewide halt on new large-load permits.
DriveNets CEO and Founder Ido Susan framed the deployment as an open, multi-vendor alternative for NeoClouds, with no lock-in to a single vendor's stack.
Sameh Boujelbene, VP Data Center Switch and AI Networks Market Research at Dell'Oro Group, said buyers are shifting from single-vendor stacks to multi-vendor clusters. She added that power availability is replacing GPU supply as the main limit on how fast capacity comes online. In her view, tuning the network and power layers together is becoming the clearest way to compete.
DriveNets and Bentaus present the joint solution at SC26 on November 19
The two companies will present the solution in the AMD booth (3028) at SC26 on Thursday, November 19, at 11:30 AM.