Remember when SSDs were getting cheaper every year and everyone predicted hard drives would disappear? Yeah, about that.
The price gap between SSDs and traditional hard drives has exploded to 16x on a per-terabyte basis, according to new data center deployment cost analysis. Not 16% more. Sixteen times more expensive.
The culprit: AI data centers are buying up NAND flash supply faster than manufacturers can build new fabs. Every major cloud provider is racing to build GPU clusters for training large language models, and those clusters need fast storage. Not cheap storage, not capacious storage—fast storage. SSDs.
NVIDIA's DGX systems, Microsoft's Azure AI infrastructure, Meta's training clusters—they're all spec'd with enterprise NVMe SSDs because training runs can't wait for spinning platters. When you're burning $100,000 per hour on GPU time, paying a premium for storage that doesn't bottleneck your training run is a rounding error.
But for normal data center operators—the ones running boring workloads like databases and file servers—this is creating a problem. SSD prices that were falling for a decade have reversed. Enterprise SSD shipments are getting diverted to high-margin AI customers, leaving everyone else fighting over scraps or paying inflated prices.
The result: hybrid storage deployments are making a comeback. Pair a small amount of SSD for hot data with a large pool of hard drives for cold storage. It's not elegant, but it's significantly cheaper than going all-SSD at current prices.
Consumer SSDs haven't been hit as hard yet—the gaming and PC market uses different product lines. But if enterprise demand keeps sucking up manufacturing capacity, don't be surprised when your next laptop SSD costs more than the last one.
The technology is impressive. The question is whether the AI boom is creating supply shocks that ripple through the entire computing industry.
NAND manufacturers are building new fabs, but those take 2-3 years to come online. Until then, expect the SSD shortage to continue. And if you're running a data center, maybe don't throw out those old hard drives just yet.
