WEKApod 3 targets AI inference with 1.1 EB storage per rack

WEKA has introduced third-generation WEKApod Nitro, WEKApod Prime, and WEKApod Prime Max appliances, positioning the rack-scale systems as a dense AI storage and memory platform for inference and agentic workloads. The appliances are designed to run with WEKA’s NeuralMesh 6 software and target data center constraints around rack space, power, and operational complexity.

Across the WEKApod 3 line, WEKA describes custom-engineered 2U chassis designs and “patent-pending” hardware changes spanning the chassis, drive interconnect, thermal management, and serviceability. The platform uses a PCIe Gen 6 internal fabric and a “cable-based, backplane-free” drive interconnect. Networking is based on NVIDIA ConnectX SuperNICs and is intended to support Spectrum-X Ethernet connectivity.

At the rack level, WEKA specifies up to 441.5 PB of raw capacity and up to 1.1 EB of “effective capacity” in a single rack with NeuralMesh data reduction enabled. The company lists throughput of 10.2 TB/s per rack and up to 210 million IOPS per rack. Those are aggressive numbers, and operators will want to validate how “effective capacity” and reduction assumptions translate to their own data and workload mix before using the figures in capacity planning.

Appliance configurations

WEKApod Nitro is described as a performance-optimized system for inference where storage bandwidth can gate GPU utilization. It uses a 2U, four-node chassis with four independent failure domains and 56 TLC drives, paired with dual-port NVIDIA ConnectX networking. WEKA lists 800 Gb/s throughput for this configuration.

WEKApod Prime is positioned as a balanced option, using a mix of AlloyFlash TLC and QLC in a 2U, four-node chassis with 56 total drives.

WEKApod Prime Max is built for maximum capacity density, using a 2U, two-node chassis with 70 NVMe drives. WEKA ties its 1.1 EB effective-capacity-per-rack figure to NeuralMesh object storage and always-on data reduction paired with Micron 245.76 TB 6600 ION drives in a 56U rack.

Thermals, failure domains, and serviceability

WEKA says the platform is engineered for 35°C ambient operation, using software-managed NVMe power throttling under thermal stress to avoid shutdown behavior. The company also calls out a “micro failure-domain” design enabled by the backplane-free architecture, intended to contain incidents to a single drive.

On serviceability, WEKA highlights hot-pluggable boot drives via a custom interposer with GUI-guided replacement, plus a mirrored NIC port layout intended to keep drive bays accessible during maintenance. WEKA characterizes a drive replacement as a “10-minute self-service operation.”

Liran Zvibel, co-founder and CEO at WEKA, said, “AI infrastructure built for traditional workloads cannot power the inference era. The constraints are different: rack space, energy, supply chain, and operational density now determine whether AI deployments produce margin or destroy it.”

WEKApod Nitro, WEKApod Prime, and WEKApod Prime Max are available to order now through WEKA’s distributor and VAR network, with delivery expected to begin in Fall 2026. More details are available on WEKA’s WEKApod product page.