Nvidia’s stock price took a hammering recently, but it’s still the big beast in the artificial intelligence (AI) hardware market. The latest buzzwords to listen out for were “agentic AI” and “reasoning” as the company announced its AI Data Platform, which had the storage suppliers trailing around after it.
Core to Nvidia’s announcements at its recent GTC 2025 event in San Jose, California, was its next-generation Blackwell Ultra graphics processing unit (GPU) for AI datacentre processing, which it says is designed for reasoning models such as DeepSeek R1 and boosts memory and inference performance.
And with Blackwell Ultra at the core, Nvidia also looked forward to a slew of rack-scale platform products in its GB/NVL line that incorporate it, plus new DGX family SuperPod clusters, workstations, network interface cards (NICs), GPUs for laptops, and so on.
This is all a bit of a pushback to the revelation that DeepSeek is more efficient and less GPU-hungry than previously seen in, for example, ChatGPT. Nvidia has used such news to assert that we’re going to need even more fast AI processing to make the most of it.
Of course, the big storage suppliers need these kinds of input/output requirements like pharmaceutical companies need disease. The requirement to process vast amounts of data for AI training and inference brings the need for storage, lots of its, and with the ability to deliver very high speeds and volumes of access to data.
So, core to announcements at GTC 2025 for storage was the Nvidia AI Data Platform reference architecture, which allows third-party suppliers – with storage players key among them – to build their kit to the GPU giant’s specs for the workloads that will run on it, that include agentic and reasoning techniques.
Those namechecked as working with Nvidia include DDN, Dell, HPE, Hitachi Vantara, IBM, NetApp, Pure Storage, Vast Data and Weka.
In slightly more detail, the announcements by these storage players around GTC included the following.
DDN launched its Inferno fast object appliance, which adds Nvidia’s Spectrum-X switch to DDN Infinia storage. Infinia is based on a key:value store with access protocols on top, but currently only for S3 object storage.
Dell announced a whole range of things, including 20-petaflop-scale PCs aimed at AI use cases. In storage, it focused on its PowerScale scale-out file system now being validated for Nvidia’s Cloud Partner Program enterprise AI factory deployment.
HPE made a big deal of its new “unified data layer” that will encompass structured and unstructured data across the enterprise, while it announced some upgrades, namely unified block and file access in its MP B10000 array.
Hitachi Vantara took the opportunity to launch the Hitachi iQ M Series, which combines its Virtual Storage Platform One (VSP One) storage and Nvidia AI Enterprise software and which will integrate the Nvidia AI Data Platform reference design, aimed at agentic AI.
IBM announced new collaborations with Nvidia that included planned integrations based on the Nvidia AI Data Platform reference design. IBM plans to launch a content-aware storage capability for its hybrid cloud infrastructure offering, IBM Fusion, and will expand its Watsonx integrations. Also, it plans new IBM Consulting capabilities for AI customer projects.
NetApp announced Nvidia validation for SuperPOD. In particular, the AFF A90 product gets DGX SuperPOD validation. Meanwhile, NetApp’s AIPod has got the new Nvidia-Certified Storage designation to support Nvidia Enterprise Reference Architectures.
Pure Storage, hot on the heels of its FlashBlade//Exa announcement, took the opportunity to reveal compatibility with the Nvidia AI Data Platform.
Vast Data launched its enterprise-ready AI Stack, which combines Vast’s InsightEngine and Nvidia DGX products, BlueField-3 DPUs, and Spectrum-X networking.
Weka announced it had achieved data store certification for Nvidia GB200 deployments. WEKApod Nitro Data Platform Appliances have been certified for Nvidia Cloud Partner (NCP) deployments with HGX H200, B200 and GB200 NVL72 products.
