Memory and SSD shortage will end in tiers, say storage vendors

WEKA and Hammerspace have issued guides on how to respond to the memory and SSD supply shortages and price hikes that now appear endemic. Both focus on tiering and on using existing media more effectively. A NetApp blog similarly offers a six-point plan centered on tiering.

DDN, Dell, Komprise, VAST Data, and VDURA have all issued suggestions about how to deal with memory and storage shortages due to AI processing demand, which prioritizes HBM for GPU memory. HBM production is seen as a necessity by memory suppliers, which reduces DRAM manufacturing capacity. Prices of both have been rising. There is a simultaneous rise in nearline storage demand by hyperscalers and enterprises, and not enough disk drives are being manufactured to satisfy it. That encourages buyers to look at high-capacity QLC (4 bits/cell) SSDs as an alternative while SSD demand is currently rising to store AI inference context memory data. Hence an SSD shortage is ongoing as well and prices are rising.

A 12-page Hammerspace Strategic Infrastructure Survival Guide says: “This is not a short-lived market cycle – it is a structural reallocation of semiconductor capacity toward components required for AI infrastructure. The practical outcome for infrastructure leaders and storage architects is a new operating reality: price volatility is persistent, and availability – not budget – can become the limiting factor for scaling storage and compute.”

Buyers should not become dependent on any single storage hardware. They should adopt a flexible, “open” data architecture allowing use of any storage media (TLC, QLC, HDD, tape) including GPU server-local SSDs. All these media types and locations, both on-prem and in the public cloud, should be brought into a single global namespace with automated data movement.

It suggests a five-point plan:

• Use the SSDs you already own as part of your shared storage
• Aggregate existing storage capacity into a shared file system
• Include cloud capacity in your global namespace
• Right-size flash usage and don’t be locked into all-flash architectures
• Extend life of existing NAS and improve asset utilization

Hammerspace includes a vendor comparison table to hammer (ahem) its points home:

We think the unnamed competitors are likely VAST Data (V), WEKA (W), and DDN (D). WEKA has also issued a NAND Flash Shortage Survival Guide:

The guide says: “Here’s the uncomfortable truth: most organizations respond to storage shortages by doing exactly what got them here – buying more storage. But when supply is constrained and prices are climbing, “just buy more” stops being a strategy and starts being a prayer.”

It makes the point that “GPU clusters waste 50-70 percent of their capacity because storage can’t feed GPUs and memory exhausts during inference.”

It recommends that “before you panic-buy more flash to hoard against the shortage, ask a harder question: Is your current infrastructure actually delivering data fast enough to keep your compute fed?”

And if it isn’t, “WEKA’s NeuralMesh fixes the architecture bottleneck. Software-defined memory co-located with GPUs delivers 90 percent-plus utilization, deploys in weeks, triples output – no new procurement.” In other words, tier across storage media intelligently, automatically and continuously, based on real access patterns.

WEKA makes a sideways dig at VAST Data by saying: “Disaggregated storage architectures require separate storage infrastructure – more servers, more NVMe, more components to source during a shortage. Software-defined approaches that deploy co-located with compute infrastructure eliminate that dependency.”

Its seven-page guide suggests you undertake a WEKA performance assessment to see how much GPU utilization you are leaving on the table.

NetApp’s blog by Product Marketing VP Jeff Baxter outlines six points in an SSD price rise and shortage action plan:

• Optimize your existing estate
• Refine data placement with tiering
• Offset storage costs with infrastructure optimization
• Expand capacity “just in time”
• Shift workloads across clouds with NetApp ONTAP
• Evaluate true TCO across the lifecycle

Several other storage suppliers have issued advice about the best way to deal with DRAM/HBM and NAND shortages and we’ve summarized them, together with Hammerspace, NetApp, and WEKA, in a table: