Nikkei’s key claim is blunt: by the end of 2027, DRAM makers will meet only 60% of demand. I only half-buy the headline framing. The snippet is too thin to tell us which market this covers: server DRAM, PC DRAM, LPDDR, or HBM-adjacent demand. Lumping all “RAM” together makes for a strong headline and a weak operating view.
The way I read this is simpler: AI has scrambled memory capex priorities. Samsung, SK hynix, and Micron have spent the last two years favoring HBM and the packaging ecosystem around it because that is where margins and customer urgency sit. Nvidia, AMD, and hyperscalers pull those investments forward. That means total DRAM bit supply can grow while specific categories stay painfully tight. We already saw that pattern in 2024 and 2025 with HBM lead times. This looks like the same pressure spreading into broader DRAM commentary.
The more useful number here is the ramp requirement: output needs to grow 12% annually in 2026 and 2027 to catch demand. That is a hard target. Memory expansion is never just “new fab announced, shortage solved.” Tools need to land, yields need to ramp, process mix matters, and qualified output always trails headline capacity. If SK’s Cheongju fab is the only disclosed 2026 production increase among the big three, that tells you the bottleneck is near-term executable capacity, not slide-deck capex.
I also have two pushbacks. First, the 60% figure is methodologically opaque in this excerpt. Is that measured in bits, revenue, or fulfillment of high-end orders? Those are very different claims. Second, the demand curve is being treated as a given. I’m not sure that is safe. If hyperscaler training buildouts cool in late 2026, or if inference efficiency keeps improving faster than expected, the gap can narrow quickly. I haven’t checked Nikkei’s full methodology, so I’d treat this as a signal that premium memory stays constrained for a while, not proof that every RAM segment faces a uniform shortage through 2030.