RA ' ANANA, Israel and Dallas, November 12 (Reuters/ PR Newswire ) – High-performance AI network solutions provider DriveNets and the energy-focused NeoCloud company Bentaus today announced a joint initiative aimed at enabling each data center GPU to generate more token, while improving efficiency per watt and operating computing power as a flexible load to help alleviate the power supply constraints faced by today's electric grids. Bentaus stated that on its clusters running AMD Instinct ™ MI325X GPU, DriveNets AI Fabric and Bentaus Ziani software, GPU power consumption has been reduced by 30%, while token throughput remains at 97%.
Bentaus develops, deploys, owns, and operates the infrastructure, hardware, and software of AI. The company has deployed DriveNets AI Fabric as the backend network for a new AI cluster, which runs AMD Instinct ™ MI325X GPU. This new cluster is used in conjunction with GPU-FLX ™ and CPU-FLX ™ within the Bentaus Ziani Systems platform to manage the computing power of enterprises, data centers, and NeoCloud in real-time. This enables the creation of an AI infrastructure that is based on open standards, responsive to power grids, and independent of suppliers, thereby maximizing the output per watt of token and reducing the cost per token.
DriveNets Ethernet-based AI fabric Improves the performance of the AMD Instinct architecture through full-stack network optimization AI From switches, NIC to RCCL and kernel optimizations. This solution is specifically optimized for Bentaus such AMD environments and is based on the reference architecture jointly designed by DriveNets and AMD. It aims to increase GPU utilization while reducing power consumption per million token, and at the same time, maximize output per second token. Published benchmark tests show that the throughput of this architecture is about 5% higher, and the first token generation time is 12% faster than publicly available results, while also minimizing power consumption and reducing effective energy costs.
The timeline for this project corresponds to the current real pressures faced by the industry: in 2026, more than a dozen states in the United States have proposed to suspend data center construction or related legislation due to concerns about the strain on the power grid and the increasing costs for electricity consumers. New York State has already implemented a statewide suspension of new large-scale load permits. Bentaus indicates that the company takes community interests into consideration when designing data centers and AI solutions. Ziani allows data centers to reduce power consumption in real-time, and by dynamically lowering GPU and CPU power consumption, it is possible to feed power back into the grid while the AI workloads continue to operate.
"Everyone is competing to increase GPU. We created Bentaus in order to increase computing power while reducing grid risks. Ziani will flexibly adjust our clusters according to the real-time grid conditions, and the fabric of DriveNets ensures that the GPU we operate does not wait idly on the network. This is how we expand the capacity of AI today: providing more computing power with higher efficiency within the power capacity that the grid can handle." said the CEO and founder of Bentaus, Bob Davidoff.
"The first round of the AI competition was about producing more token. The next round of the competition is about producing them with less electricity. With the help of AMD and Bentaus, we have come up with a solution: a complete, open system. Our Ethereum fabric allows each GPU to achieve higher utilization and more token per watt, while the Ziani of Bentaus ensures that this computing power operates above the level of electricity that the power grid can handle. This is the open, multi-supplier alternative that NeoCloud has been looking for – providing top-notch performance-to-power consumption ratios without locking in to a single supplier's technology stack." said the CEO and founder of Ido Susan, DriveNets.
"Two changes are simultaneously reshaping the infrastructure of AI. Buyers are shifting from a single supplier technology stack to a multi-supplier cluster, and power availability is replacing GPU supply as the main limitation to the speed at which capacity can be added. What stands out about this deployment is that the network layer and the power layer are being optimized together to extract the most token from each GPU and every watt of power. This is quickly becoming the most clear competitive approach," said Dell, Vice President of Data Center Exchange and AI Network Market Research at Oro Group.
Understanding DriveNets and Bentaus:
- At 11:30 a.m. on Thursday, November 19th, participate in the introduction of the joint plan by DriveNets and Bentaus at the SC26 conference at booth AMD (3028).
- Read the Bentaus case study
- Visit the official websites of DriveNets, drivenets.com, Bentaus, and bentaus.com.
About DriveNets
DriveNets is a leader in large-scale network solutions for service providers and AI infrastructure. The company has created a new way of building networks, significantly improving the economic model of networks and optimizing their utilization and efficiency. DriveNets AI Fabric was launched in 2023, offering high-performance Ethernet AI networks and is one of the best alternatives to Ethernet outside of InfiniBand. This solution has been adopted by global hyperscale cloud service providers, NeoCloud, and enterprises. For more information, see drivenets.com.
About Bentaus
Bentaus is an organization that focuses on energy efficiency. It develops, deploys, owns, and operates the infrastructure, hardware, and software of AI. Through the energy orchestration platform Ziani Systems tailored for enterprises, data centers, and NeoCloud, Bentaus can manage computing power and power consumption in real-time intelligently. This platform supports flexible and grid-responsive AI infrastructure, enabling maximum token output while enhancing performance, energy efficiency, and the cost-effectiveness of computing power. At the same time, it minimizes energy consumption and the costs associated with each token. Bentaus relies on open-source technologies and industry standards to provide AI solutions that are independent of suppliers, with the aim of maximizing token production and minimizing energy consumption as well as the costs of each token.
