Bentaus deploys DriveNets AI fabric and AMD GPUs, claims 30% less energy per token generated

Bentaus, which describes itself as an energy-first NeoCloud, has deployed DriveNets’ high-performance AI fabric and AMD GPUs. The announcement, dated Nov. 13, 2026, says the deployment will reduce the energy required to generate AI tokens by 30%.

Bentaus is touting a 30% reduction in the energy required to generate AI tokens and has identified AI networking infrastructure, in addition to GPUs, as part of the deployment. When assessing NeoCloud efficiency, it is important to distinguish the company’s stated energy-per-token figure from the power consumption of the entire facility.

The focus of the announcement is not GPU performance alone, but the evaluation of a configuration combining computing resources and AI networking in terms of energy per token generated. Bentaus describes its NeoCloud as “energy-first” and has deployed DriveNets’ fabric and AMD GPUs. In other words, the announcement promotes an integrated infrastructure that includes data transfer as well as computing. However, the 30% figure is the company’s claim, and the information provided does not specify the comparison baseline or workload conditions. This metric should be considered separately from the facility’s overall power consumption. Even if energy per token falls, an increase in operating volume does not necessarily mean total energy consumption will fall by the same proportion. An assessment therefore requires a comparison using consistent models, workloads, measurement periods and baseline configurations. The excerpt does not identify the GPU model or number of units, the fabric configuration or the measurement method, so it is not possible to determine whether the 30% figure applies to other environments. Related stories from Physical AI News include AMD announces start of shipments for sixth-generation EPYC, which covers CPU shipments, and Generative Bionics announces construction of a European “physical AI data factory” for humanoids, adopts AMD platform, which covers computing and data infrastructure for physical AI. These are separate announcements and do not describe the same products or operations as the Bentaus configuration. However, because AMD appears in different computing-infrastructure contexts, performance comparisons should distinguish among products and applications. The industry implication is whether cloud providers can use the efficiency of an entire system, including networking, as a selling point alongside computing performance. Points to watch include the measurement conditions behind the 30% reduction, utilization and total power consumption in actual operations, and the specific GPU and network configuration. Until these details are available, the announced reduction cannot be treated as a benchmark for comparisons with other companies’ environments.

The announcement describes DriveNets as a provider of high-performance AI networking solutions and Bentaus as a NeoCloud. The excerpt provided does not identify the GPU model or number of units, the network configuration, or the comparison baseline and measurement conditions for the 30% reduction.