Dell AI server revenues leap but storage waits on Project Lightning
During the past quarter, Dell’s datacentre sales went through the roof due to new server and networking products for AI – but storage has not shared the same dynamic
Dell’s quarterly results show a huge growth in server sales, driven by artificial intelligence (AI) projects, but a relative lag in storage. Key reasons behind that might be that Dell’s current storage lags a little behind the curve in AI performance, while its massive parallel network-attached storage (NAS) that aims to plug that gap, Project Lightning, is in gestation.
Dell’s PC division usually massively outsells its datacentre products, but that’s not the case in its latest (second) quarterly results, which show 69% growth in sales of servers and networking equipment year-on-year. That equated to a revenue of $16.8bn for the infrastructure division that put the client services – i.e. personal equipment – into the shade with $12.5bn of sales.
Dell has benefited here from being the first to the AI market, with servers, the latest Nvidia graphics processing unit (GPUs), and switches compatible with high throughput Nvidia Spectrum-X networking and Ultra-Ethernet cards.
“In the last six months we have delivered $10bn worth of servers for AI,” said Jeff Clarke, vice-chairman and chief operating officer for Dell Technologies. “That’s more than was attained in the whole previous year. Demand is strong and sales of the new AI hardware has totalled $20bn for the year.”
Overall, Dell’s Q2 results showed record revenue of $29.8bn, which was up 19% on the previous year. Of the $16.8bn of revenue contributed by the infrastructure group – growth of 44% in a year – servers and network equipment contributed $12.9bn.
Meanwhile, however, storage arrays – flash and disk – saw revenues lower by 3% over the year at $3.9bn in the quarter.
Meanwhile, the client services group’s revenue growth was a mere 1% year-on-year, with enterprise PCs reporting $10.8bn revenue (+2%) and consumer products $1.7bn (-7%).
Storage the poor relation in infrastructure sales
A salient feature of these results is that the demands of AI seem to favour compute hardware more than storage.
That might be confirmed by the latest results from NetApp, which is number one in flash storage arrays, according to IDC. Here, the array maker posted quarterly results of $1.56bn in August, which equated to annual growth of 1%.
Meanwhile, Pure Storage announced revenue of $861m, and that was an increase of 13% on sales in a year – but there’s a catch. That set of figures included its delivery – unprecedented – of SSD DirectFlash Modules (DFM) to hyperscaler Meta.
Pure’s DFMs are a proprietary format in which the vendor has packed a much higher density of storage onto SSD cards. That’s because it offloads a lot of on-board cache to the array and handles data there instead.
HPE’s third quarter results showed revenue growth for servers at $4.9bn, up 16% year-on-year, but doesn’t appear to break out storage revenue.
Towards evolution in storage
Why have we seen a boost in revenue for servers for AI, but not really with storage?
There’s no doubt from a technical point of view that storage is an essential support for compute for AI; it’s possible that enterprises have staged their budget spend and focused first on processing power.
At the same time, it’s true that storage products have lagged behind in terms of performance compared with compute. For example, servers that feed GPUs are able to move data at a rate of 400Gbps or even 800Gbps. Current storage products offer around 100Gbps.
Storage suppliers have, however, centred efforts to develop AI storage around parallel file system storage for AI.
Vast Data led the way here, with massive parallel access to storage, while Hammerspace and Weka also followed.
Dell responded with Project Lightning – which comprises Powerscale, the rebranded Isilon scale-out NAS – but that doesn’t seem to have a release date yet. Meanwhile, NetApp has Ontap Data Platform for AI, while Pure has FlashBlade//Exa.
Read more about storage for AI
- Storage technology explained: AI and data storage. In this guide, we examine the data storage needs of artificial intelligence, the demands it places on data storage, the suitability of cloud and object storage for AI, and key AI storage products.
- AI storage: NAS vs SAN vs object for training and inference. Artificial intelligence operations can place different demands on storage during training, inference, and so on. We look at NAS, SAN and object storage for AI and how to balance them for AI projects.
