MSI is warning that GPU shipments will drop by 20% due to a shortage of high-bandwidth memory. Not chip shortages, not manufacturing capacity—RAM. Specifically, the specialized memory that high-end graphics cards need to function.

This is supply chain chaos in microcosm: a single component bottleneck cascading through an entire industry. And it's happening at possibly the worst time, when AI training, gaming, and graphics workstations are all competing for the same limited production capacity.

Here's what's going on: Modern GPUs don't just need processing cores—they need massive amounts of extremely fast memory right next to those cores. GDDR6 and HBM (High Bandwidth Memory) are produced by only a handful of manufacturers, primarily Samsung, SK Hynix, and Micron. When demand spikes beyond their production capacity, everyone downstream suffers.

The AI boom is the primary culprit. Data centers are buying GPUs by the tens of thousands for training large language models. NVIDIA's H100 and H200 chips—each using enormous amounts of HBM—are sold out months in advance. That production is soaking up memory supply that would otherwise go to consumer graphics cards.

MSI's 20% figure is probably conservative. Other manufacturers are likely facing similar constraints but haven't gone public yet. This means gamers and professionals looking to upgrade their systems will face higher prices, longer wait times, or both.

There's a grim irony here: the technology revolution everyone's excited about is being constrained by manufacturing capacity for commodity components. We can design chips with billions of transistors, but we can't produce enough of the memory to feed them.

Memory manufacturers aren't sitting idle. SK Hynix and Samsung are both ramping up HBM production, but building new fabrication capacity takes years and billions in investment. Short-term relief is unlikely. We're looking at constrained supply through at least 2027.

The gaming industry is particularly frustrated. Console manufacturers, GPU makers, and PC builders are all competing for the same memory chips. Gamers accustomed to steady hardware upgrades are discovering that AI training has jumped to the front of the line—and it's willing to pay premium prices that consumer products can't match.

This shortage is different from the 2020-2021 chip crisis. That was about fabrication capacity and COVID-related disruptions. This is about a specific, high-tech component where there are few manufacturers and no easy substitutes. You can't just swap in different memory—GPUs are designed around specific memory architectures.

Some manufacturers are exploring workarounds: using slightly older memory standards, redesigning boards to be more memory-efficient, or allocating production toward lower-end cards that use less demanding memory. None of these are ideal solutions—they're just ways to ship something while waiting for supply to catch up with demand.

The situation will eventually resolve. Memory production will increase, demand will stabilize, and supply chains will rebalance. But "eventually" doesn't help someone trying to buy a GPU in 2026.

For now, expect graphics cards to remain expensive and scarce. The AI boom has consequences that ripple far beyond data centers and chatbots. One of those consequences is that building a gaming PC just got a lot more frustrating.

The technology is incredible. The supply chain is not. And right now, the bottleneck isn't chips—it's the memory to feed them.