AI Data Centers Are Eating the World's Memory Chip Supply
Global chip sales just hit the highest monthly total ever recorded. The reason your next laptop, console, or phone costs more has everything to do with it.
If you've priced out a new laptop, gaming console, or server lately and flinched, you're not imagining it. The global semiconductor market posted $120.6 billion in sales in May 2026, according to the Semiconductor Industry Association's World Semiconductor Trade Statistics report - up 9.2% from April and a staggering 104.1% higher than May 2025. It was the highest-ever recorded monthly total, and the 15th straight month of month-over-month growth. "The global semiconductor market continued to grow substantially in May, hitting the highest-ever recorded monthly sales total," SIA president John Neuffer said in the report.
That headline number sounds like unambiguous good news for the chip industry. But underneath it is a much messier story, and it's one that's now landing directly on ordinary consumers: the AI industry's appetite for memory has broken the market for RAM.
The squeeze
Three companies - Samsung, SK Hynix, and Micron - control more than 95% of the world's DRAM production. For the past two years, that capacity has been getting redirected. High-bandwidth memory, the specialized, stacked chips that feed data-center AI accelerators, now carries margins several times higher than ordinary DDR5. Industry estimates suggest HBM's share of global DRAM wafer output has roughly tripled since 2024, and AI data centers are now believed to absorb something like 70% of all high-end memory output worldwide, up from an estimated 20-30% before the AI buildout accelerated.
Every wafer a manufacturer commits to HBM for a hyperscaler's next data center is a wafer that isn't making the DDR5 sticks that go into laptops, desktops, and game consoles. SK Hynix has said its entire 2026 HBM production was already fully committed as of last fall. Micron CEO Sanjay Mehrotra has told investors to expect supply tightness to continue into 2027. That's the mechanism: not a factory fire or a trade dispute, but a deliberate, margin-driven reallocation of a scarce resource toward whoever is willing to pay AI-era prices for it.
What it costs now
The price effects have been sharp and fast. Server-grade 64GB DDR5 RDIMM pricing has roughly doubled between Q1 2025 and Q1 2026. Consumer DRAM contract prices climbed by similarly dramatic margins in single quarters this year. The pain has shown up in finished products, too: Sony's PlayStation 5 Digital Edition rose from $499.99 to $599.99, Nintendo's Switch 2 went from $449.99 to $499.99, and Microsoft's Xbox Series S climbed from $399.99 to $499.99. A high-end Steam Deck OLED that cost $549 is now closer to $789.
A general manager at memory retailer Team Group put it bluntly earlier this year: DRAM and NAND prices had "doubled in a single month," and the pricing crisis "has only just started."
No consensus on when it ends
What's striking is how little the industry's own leaders agree on a timeline back to normal. SK Hynix's leadership has floated the possibility that tight supply persists past 2030. Micron points to 2027. Intel has signaled meaningful relief is unlikely before 2028. When the companies actually making the chips can't converge on an end date, it's a reasonable bet that this isn't a short-term inventory hiccup - it's a structural shift in who gets first claim on the world's advanced memory manufacturing.
The bigger picture
Most of the AI boom's costs have stayed relatively abstract to the average person: data center power consumption, GPU shortages measured in the tens of thousands of units, capital expenditure numbers with too many zeros to feel real. The memory crunch is different. It shows up as a real number on a receipt, whether you're buying a laptop, upgrading a home server, or just replacing a game console.
That makes the memory shortage a useful stress test for a question the AI industry has mostly avoided answering directly: when compute infrastructure and consumer hardware compete for the same finite manufacturing capacity, who wins by default, and is that the outcome anyone actually chose - or just the one the market backed into? For now, the wafers are voting with their margins, and the answer is AI, every time.
Sources
Semiconductor Industry Association, Global Semiconductor Sales Increase 9.2% Month-to-Month in May: https://www.semiconductors.org/global-semiconductor-sales-increase-9-2-month-to-month-in-may/
Tom's Hardware, DRAM chip supply to module makers could drop by more than 70% year-on-year in 2027: https://www.tomshardware.com/pc-components/ram/dram-chip-supply-to-module-makers-could-drop-by-more-than-70-percent-year-on-year-in-2027-says-apacer-ceo-demand-for-hbm-and-server-ram-continues-to-devour-manufacturing-capacity
Tom's Hardware, The RAM pricing crisis has only just started, Team Group GM warns: https://www.tomshardware.com/pc-components/dram/the-ram-pricing-crisis-has-only-just-started-team-group-gm-warns-says-problem-will-get-worse-in-2026-as-dram-and-nand-prices-double-in-one-month
Everstream Analytics, Global Memory Chip Shortage Worsens: https://www.everstream.ai/risk-centers/global-memory-chip-shortage-worsens/
Shattered.io, RAM Prices Up 89%: AI Memory Crunch Hits Gaming: https://shattered.io/ram-prices-ai-memory-shortage-2026/
IEEE Spectrum, How and When the Memory Chip Shortage Will End: https://spectrum.ieee.org/dram-shortage