WEKA has struck a collaboration agreement with Nebius that will see the high end storage purveyor’s platform made available through the GPU-as-a-service vendor’s datacenters.
Netherlands-based Nebius focuses on supporting AI workloads, which are naturally data hungry. The firm already has agreements with other high-end storage providers – it struck a deal with DDN back in May.
However, WEKA said that its tech would be offered as “Nebius’ highest performance data storage solution, designed for AI innovators across multiple industries who have extremely performance-intensive workloads.”
The firms said the partnership would enable Nebius’ customers to “scale compute and storage resources on demand with ultra-high performance and microsecond latency for efficient AI model training and precision AI inference.”
In addition, WEKA said, “Customers with deployments ranging in scale from petabytes to multiple exabytes of data stored, requiring hundreds to tens of thousands of GPUs, are also supported.”
In practice, that could mean “AI model providers driving real-time AI agent responses and scientific institutes accelerating critical research discoveries.”
WEKA’s software will be run natively on Nebius’ server hardware, and will be fully integrated into its AI Cloud platform stack. Critically, the tie-up should keep those hideously expensive, power guzzling GPUs saturated with data, which WEKA claimed would maximise both performance and power efficiency. Estimates of the amount of time GPUs are idling during training runs range as high as 57 percent thanks to data transfers.
“WEKA offers exceptional levels of throughput, metadata and IO performance, at consistently low microsecond-level latency — capabilities that legacy storage solutions simply can’t provide,” the storage firm claimed.
Global ambition
The offering will span Europe and the US initially, but will be rolled out globally. Nebius only recently opened its first UK operation, deploying a complement of Nvidia Blackwell Ultra GPUs.
That represented its seventh cluster worldwide, with operations in the power rich Iceland and Iceland, as well as France, Israel, and two sites in the US.
The firm unveiled its AI cloud platform just last October. It specced plans to offer NVIDIA H100 and H200 Tensor Core GPUs, L40S GPUs, and the Nvidia GB200 NVL72, and promised high-speed storage delivering up to 100 GBps and 1 million IOPS.
This all followed a resumption of trading its shares on NASDAQ. CEO Arkady Volozh was a cofounder of Russian tech giant Yandex. The holding firm was listed on NASDAQ and HQ’d in Amsterdam. But the listing was suspended following Russia’s invasion of Ukraine. A subsequent restructure saw it offload the Russian operations, restructure as Nebius, and launch itself into the GPU as a service market.
