AI is turning the memory market from a familiar boom-and-bust cycle into something a lot less forgiving. Jefferies expects DRAM and NAND shortages to ease only by 2028, which means the old ritual of a price spike followed by a comfortable dip may be gone for good.
The near-term numbers are ugly for buyers. Contract memory prices are expected to rise by 40% to 50% this quarter, then climb another 30% to 40% in the fourth quarter. Jefferies also sees prices continuing to rise in 2027, with a further increase of 40% to 45% over the period, as manufacturers simply do not bring enough new capacity online fast enough.
Memory prices rise as cloud buyers take the best of supply
The market is being pulled apart by who can sign the biggest contracts. Cloud giants already absorb up to 50% of available memory supply, and that share could reach 70%, helped by long-term deals that run for at least two years and often include advance payments of up to 40% of the purchase price. Consumer electronics makers do not get that kind of treatment, which is a polite way of saying they are left fighting over what is left.
That is a bad setup for everything from phones to laptops, because weaker access to memory does not just push up bills – it can cap production volumes. In other words, this is not a tidy pricing story. It is a supply-allocation story, and the biggest AI spenders are winning it.
HBM keeps eating the DRAM factory floor
Another reason the shortage is sticky: the memory makers are chasing the profitable stuff. High-bandwidth memory, or HBM, keeps taking more space on production lines, which leaves less room for classic DRAM. Samsung Electronics is still trying to close the gap with rivals as it moves toward HBM4, after trailing them in HBM3 and HBM3E.
That shift is not new, but it is getting more expensive for everyone else. Once a manufacturer can earn more from AI-focused memory than from standard chips, the consumer market becomes the side project – and the side project gets rationed.
China will help NAND before it helps DRAM
Chinese makers are not in a position to fix the global picture anytime soon. Jefferies says they are still one and a half to two generations behind Western and Korean competitors, and that gap means they are unlikely to change the DRAM market before the end of 2027. CXMT, for example, is seen as constrained without EUV lithography and unlikely to move to DDR6 or HBM3E in the foreseeable future.
NAND is a different story. By 2028, Chinese suppliers are expected to be much closer to global levels in both technology and volume, which could give that segment more breathing room than DRAM. Jefferies says DRAM prices would need global output to rise by 15% to 20% by 2028 before they can start falling, and even that depends on AI demand finally slowing down. That is a big ”if” in a market currently being fed by data centers with very deep pockets.
So the likely path is simple enough, if not especially comforting: memory stays tight, AI keeps getting first dibs, and mainstream buyers keep paying the bill. The real question is whether new supply can outrun demand before 2028, or whether the industry has already decided that permanent scarcity is the new normal.