Wonjin Lee, Samsung’s president of global marketing, didn’t hedge when he addressed what’s coming for the memory chip market: “In 2026, there’s going to be issues around semiconductor supplies, and it’s going to affect everyone, not just Samsung.” That’s a striking thing for a company that makes a meaningful share of the world’s DRAM and NAND flash to say out loud, and the pricing data backs him up. Samsung raised the retail price of a 32GB DDR5 memory module from $149 to $239 last September — a 60% jump — and contract DRAM pricing has since surged past 100%, from roughly $7 to about $19.50 per unit. Prices are already up around 50% year-to-date, with Gartner forecasting a 47% full-year increase for 2026 overall, and some 64GB server memory modules could cost double their early-2025 price by the end of this year.
This isn’t a temporary supply hiccup working its way through one bad quarter. SK Hynix, one of the three companies that together make almost all of the world’s memory chips, says its DRAM, NAND, and high-bandwidth memory production capacity is essentially sold out for the entirety of 2026. Micron has reportedly exited the standard consumer memory market altogether to concentrate on higher-margin product lines. And SK Hynix’s own internal expectation, according to reporting on the shortage, is that it could persist well beyond 2030 — not months, years.
AI Data Centers Are Eating the World’s Memory Supply
The mechanism behind the shortage is straightforward once you see it: high-bandwidth memory, the specialized chip stacks that feed AI accelerators inside data centers, consumes roughly three times the silicon wafer capacity per gigabyte compared to standard consumer DRAM. Hyperscale cloud providers — Meta, Google, Microsoft, and Amazon among them — have locked in long-term supply agreements directly with memory manufacturers to guarantee they get enough HBM to keep building out AI infrastructure. Every wafer of manufacturing capacity that goes toward satisfying those contracts is a wafer that isn’t going toward the DRAM that ends up inside a laptop, a phone, or a family’s desktop PC. The AI buildout and the consumer electronics supply chain are now directly competing for the exact same physical manufacturing capacity, and the AI buildout is winning, because it pays more per wafer.
The most recent quarterly numbers show that squeeze playing out in real time. SK Hynix’s DRAM pricing rose roughly 30% sequentially last quarter, and Samsung’s rose more than 40%. NAND flash pricing climbed even faster — mid-50% growth at SK Hynix, high-60% at Samsung. On the mobile side specifically, Samsung and Micron reportedly pushed for price increases above 80% compared to the first quarter of the year, while SK Hynix’s opening ask was a comparatively restrained 55% to 60%. Those aren’t abstract industry statistics. They’re the input costs that device makers either absorb into thinner margins or pass straight through to a retail price tag.
Not Quite as Simple as “Prices Keep Climbing Forever”
It’s worth being honest about a genuine wrinkle in this story rather than flattening it into a single panic narrative. The most recent earnings data from SK Hynix and Samsung actually came in below what Wall Street analysts had modeled — SK Hynix’s 30% sequential DRAM increase, for instance, was well under the roughly 39% analysts expected. Part of the explanation is that customers are increasingly locking in multi-year supply agreements at prices below the current market peak, in exchange for guaranteed allocation. That stabilizes revenue for the memory makers and gives device manufacturers some predictability, but it also means the sharpest pricing pain from this shortage may be smoothed out across several years of contracts rather than delivered all at once in a single brutal price spike. It’s a real shortage with real, already-visible price effects — it just isn’t a straight line upward from here, and anyone treating it as one is oversimplifying a genuinely more complicated market.
What isn’t in dispute is the demand-side effect already showing up on device makers’ own forecasts. Apple and other major manufacturers have signaled that the memory shortage will constrain 2026 production, with analysts projecting global smartphone shipments could decline nearly 13% this year and PC shipments could contract more than 11%, as manufacturers either build fewer units or push retail prices higher to protect margins on the memory they can actually secure.
What This Means for Philippine Founders
The Philippines isn’t just a downstream price-taker in this story — it’s a genuine node in the physical supply chain, with a substantial share of the country’s manufacturing exports tied to semiconductor and electronics assembly and testing. A multi-year memory crunch cuts two ways here: it’s real upstream cost pressure for any Philippine business that touches hardware, but it’s also a live opportunity, as global manufacturers look to diversify assembly and test capacity away from concentrated risk in a handful of markets — worth watching closely for anyone building in or around the country’s existing electronics manufacturing base.
For consumer-facing Philippine startups specifically — device financing platforms, refurbished and secondhand phone marketplaces, telco device bundling programs, school and enterprise laptop procurement — this isn’t a distant global headline. It’s a dateable cost-input shock that will land in supplier quotes over the next two to four quarters, whether or not it makes local news. Founders modeling 2027 unit economics should be stress-testing hardware and device costs against these numbers now rather than assuming today’s supplier pricing holds. And for the Philippines’ growing bench of AI-adjacent startups, the underlying resource fight here is worth watching as a preview: as hyperscalers keep locking up memory and compute capacity years in advance, any Philippine founder planning to eventually run their own inference or training infrastructure — rather than renting it from a cloud provider — needs to budget for genuinely scarce, genuinely expensive hardware, not the commodity pricing that existed even two years ago.
Share this article