Samsung Electronics, SK Hynix and Micron make more than 95% of the world’s DRAM. They have now moved roughly 93% of their combined manufacturing capacity to high-bandwidth memory, the specialised chips inside AI accelerators such as Nvidia’s GPUs.
The reason is money. HBM is estimated to earn three to five times as much per wafer as conventional DRAM. Output for phones, laptops and televisions has been squeezed to make room, and the supply of consumer-grade DRAM and NAND flash has cratered.
Where consumers feel it
Memory now takes a bigger bite out of a television’s cost of materials: 6-7%, up from 2.5-3%. Makers working on thin margins tend to pass that on at the till.
TrendForce projects notebook shipments will fall 9.4% year over year in 2026. It expects the average smartphone price to reach a record $523. At that level a good slice of the global market gets priced out, and the mid-range handsets that move volume in Southeast Asia, India and Latin America feel it most.
No quick end
Micron chief executive Sanjay Mehrotra has predicted the shortage will run well beyond 2026. SK Group chairman Chey Tae-won has gone further and suggested it could last until about 2030.
SK Hynix plans to double its wafer capacity over the next five years. That extra capacity will probably go to the same high-margin HBM products that caused the squeeze. The company has been the most aggressive of the three in chasing HBM demand, and the expansion suggests it sees no ceiling on AI orders soon.
Manufacturers are reportedly moving to shorter contracts and tighter order management to curb hoarding. Buyers who stockpile out of fear of shortages make the shortage worse.
The strategic picture is lopsided. Three companies control the market, sell their most profitable product in growing volumes, and can see demand years ahead. Mehrotra’s public comments amount to telling investors to expect strong pricing power for the foreseeable future. Everyone downstream faces rising component costs and no obvious relief.
