WEKA brings NeuralMesh AI file system to Nvidia BlueField-4 DPUs

The WEKA NeuralMesh high-performance AI file system software will be able to run in Nvidia’s forthcoming Bluefield-4 DPUs, enabling flash JBODs fronted by BlueField with no x86 controller processor involved at all.

A DPU is a data processing unit, ARM-powered hardware that accelerates storage processing functions, originally designed to offload a host server/controller x86 CPU. But now the latest Nvidia BlueField-4 DPU, with its Grace CPU and ConnectX-9 networking, is so powerful, with 6 times more processing power than BlueField-3 and supporting 800 Gbps throughput, it can run storage controller and file system functions faster than an x86 CPU. BlueField-4 also has multi-tenant networking, rapid data access, AI runtime security, and native support for Nvidia DOCA microservices. So storage system suppliers can ditch their x86 controller/server and run their software entirely on the DPU, as WEKA is doing with its NeuralMesh parallel file system.

WEKA co-founder and CEO Liran Zvibel said: “WEKA’s NeuralMesh architecture, built for Nvidia BlueField-4, is a tremendous leap forward, providing the foundational storage for next-generation AI factories.” It will simplify “AI infrastructure deployment while dramatically improving economics, performance, and power efficiency.”

Nilesh Patel, chief strategy officer at WEKA, explained: “Power efficiency is key to bringing gigascale AI factories online. Nvidia BlueField-4 and WEKA’s NeuralMesh storage software will directly address this challenge, delivering breakthrough tokens-per-watt efficiency without compromising performance, scale, or operational simplicity. We view NeuralMesh’s integration with BlueField-4 as a strategic growth engine that can simplify and accelerate Enterprise AI factory deployments for our customers.”

WEKA says NeuralMesh with BlueField-4 will deliver a greater than 100x tokens/watt improvement for agentic AI workflows compared with traditional, CPU-attached storage systems.

The company is a validated Nvidia AI Data Platform reference design partner.

The company is developing its next-generation NeuralMesh software release, along with NeuralMesh Axon and Augmented Memory Grid for BlueField-4, with detailed roadmap information to be shared with customers and partners in early 2026. Read more in a WEKA blog.