The price of a 32GB DDR5 memory module went from $94 to $282 in a single quarter. That’s a 122% increase between Q3 2025 and Q1 2026, and the culprit isn’t some exotic supply chain disruption or a factory fire. It’s Big Tech writing very large checks.
US hyperscalers, the companies that operate the world’s largest data centers, have been placing enormous orders for memory chips to feed their AI infrastructure buildout. Google, Amazon, Microsoft, and Meta are all in the mix, and their appetite is so voracious that some agreements now account for up to 40% of global DRAM output.
The AI memory vacuum
Memory chips come in two main flavors that matter here: standard DRAM, which goes into everything from laptops to cars, and high-bandwidth memory (HBM), the specialized stuff that AI accelerators crave. Both are in short supply, but for different reasons.
Samsung, SK Hynix, and Micron, the three companies that dominate global memory production, have been redirecting their manufacturing capacity toward high-margin AI products. That means HBM and server-grade DRAM get priority, while the conventional chips destined for PCs, smartphones, and other consumer electronics get pushed to the back of the line.
Analysts are describing the current environment as a “structural supercycle,” and expectations are that these shortages will persist at least until 2028.
Consumers are already paying the price
Apple raised prices on its products by nearly 20% in June 2026, directly citing elevated memory costs. Microsoft bumped Xbox prices by at least $100. These aren’t subtle adjustments buried in spec sheets. They’re the kind of increases that show up on a credit card statement and make people reconsider purchases.
And it’s not just gadgets. The same memory chips that power your laptop also go into automobiles, medical devices, and industrial equipment. When hyperscalers absorb a disproportionate share of global supply, the shortage cascades across sectors that most people don’t associate with Silicon Valley’s AI ambitions.
A trillion-dollar forecast and Washington’s attention
The scale of what’s happening becomes clearer when you look at the revenue projections. Memory sales stood at roughly $230 billion in 2025. Forecasts suggest that figure could surpass $1 trillion by 2027. That’s more than a fourfold increase in two years, driven almost entirely by AI and data center demand.
The sheer magnitude of the shift has drawn attention in Washington. Lobbying efforts have intensified, with proposals on the table that include invoking the Defense Production Act, a wartime-era law that allows the federal government to direct private companies to prioritize certain orders.
The hyperscalers themselves face a different calculus. Hundreds of billions of dollars are flowing into AI infrastructure in 2026 alone, and memory is one of the critical bottlenecks. For companies racing to build out AI capabilities, paying a premium for guaranteed supply is a rational decision, even if it distorts the broader market.