Remember when you could grab a decent 1TB SSD for under $50? Those days are over, and you can thank AI datacenters for that.
Kioxia, one of the world's largest NAND manufacturers, just confirmed what PC builders have been watching in horror: their entire supply is sold out for 2026 and likely through 2027. A company executive stated bluntly that "the era of the cheap 1TB SSD is over."
The culprit? AI infrastructure is eating the world's NAND supply.
Datacenters training large language models and running inference need massive amounts of fast storage. When Microsoft, Google, Meta, and Amazon are buying flash storage by the shipping container, consumer SSD manufacturers get whatever's left over. And what's left over costs more.
This is a concrete example of how AI infrastructure spending affects regular people. You're not training models. You're not running inference. You just want to upgrade your laptop or build a gaming PC. But you're competing with companies that have effectively infinite budgets and are in an arms race to deploy more compute.
NAND manufacturers love this, of course. Why sell to consumers at thin margins when you can sell bulk contracts to hyperscalers at premium prices? The economics are obvious. The consumer market is getting priced out because there's a much more profitable customer willing to pay more.
Price increases are already showing up. SSDs that cost $60 six months ago now run $90-$100. 2TB drives that were approaching affordable are back above $150. And according to Kioxia, don't expect relief until late 2027 at the earliest.
Here's the part that really frustrates me: this was predictable. When the AI boom started, anyone looking at NAND supply chains could see this coming. But nobody in the industry warned consumers, because why would they? Higher prices mean better margins.
We're seeing this pattern across the hardware stack. GPUs? Impossible to buy at MSRP because datacenters buy entire production runs. High-bandwidth memory? Same story. Now it's spreading to SSDs, and soon it'll hit DRAM.
The AI boom is real. The technology is impressive. But the infrastructure buildout is creating shortages and price increases for components that have nothing to do with AI training. You're paying more for storage because needs another datacenter.
