If you’ve been putting off a RAM upgrade because prices have become ridiculous, there is finally a little light at the end of the tunnel. Industry forecasts suggest DRAM supply could move ahead of demand in 2028, potentially creating the conditions for prices to fall.
That’s after an exceptionally expensive 2026 and what could be an even tighter 2027. The catch is that memory manufacturers have learned from previous boom-and-bust cycles, so additional supply may not translate into an immediate bargain-bin bonanza.
Manufacturers chose HBM margins over your DDR5 stick
Slower price increases still aren’t the same as cheaper RAM
The current problem is not simply that the world suddenly ran out of ordinary desktop memory. The AI infrastructure boom changed what memory manufacturers want to produce. Samsung, SK Hynix, and Micron can earn more by dedicating advanced production capacity to high-bandwidth memory, or HBM, and server products than by flooding the market with inexpensive consumer DRAM. That shift has restricted the supply available for desktops, laptops, phones, and other mainstream consumer devices, even while consumer demand has remained relatively low.
The resulting effect on prices has been dramatic, to say the least. TrendForce forecast conventional DRAM contract prices would rise 90 to 95 percent quarter over quarter in the first quarter of 2026, with PC DRAM prices expected to more than double. By the third quarter, the firm expected the increase to slow to 13 to 18 percent, but that still represents another rise rather than a reversal. Supplier inventories also remained at historically low levels, while most additional output was being directed toward servers.
In other words, slower inflation is not the same thing as cheaper RAM. Prices can rise less aggressively while a 32GB kit remains painfully expensive. For anyone waiting to build or upgrade a PC, the meaningful turning point is when production growth consistently catches up with demand, not when quarterly price increases merely get smaller.
2028 is when supply might finally overtake demand
New fabs take years to build, qualify, and ramp
The turning point for supply may arrive in 2028. A TrendForce projection presented at the 2026 Future of Memory and Storage conference indicated that DRAM supply could exceed demand by about 2% that year. JP Morgan’s modeling points in a similar direction, estimating that DRAM supply growth could beat demand growth by roughly 3 percentage points in 2028 after substantial shortfalls in 2026 and 2027.
There is an important reason the relief takes so long. New fabs cannot be switched on like extra servers in a data center. Manufacturers must construct cleanrooms, install complex equipment, qualify production processes, improve yields, and then ramp volume. TrendForce says capacity scheduled to come online in 2027 is unlikely to make a substantial contribution until 2028. Until then, AI servers will continue competing with conventional products for a limited pool of DRAM wafers.
Once supply overtakes demand, however, buyers regain some leverage. PC manufacturers and module vendors no longer need to fight as aggressively for allocation, inventories can rebuild, and suppliers face more pressure to negotiate. Bernstein reportedly expects SK hynix’s average DRAM selling price to peak at $2.23 per GB near the end of 2027 before dropping to $1.05 per GB by the end of 2028, a decline of more than 50%.
A surplus on paper doesn’t guarantee a discount at checkout
Manufacturers have every incentive to keep margins fat
All of this sounds like excellent news, except it doesn’t necessarily mean retail RAM will instantly become cheaper again. A forecasted industry surplus is small, and it could disappear if AI demand grows faster than expected, new factories ramp slowly, or manufacturers adjust production. Citi, for example, has offered a much more aggressive demand forecast that keeps DRAM undersupplied in both 2027 and 2028. These are competing models, not promises.
More importantly, the three dominant memory manufacturers have little incentive to recreate the enormous gluts that historically crushed DRAM prices. Samsung has said it intends to balance capital spending with demand and pricing to reduce oversupply risk. The industry is also focusing on process migrations and high-margin products instead of rushing to add conventional wafer capacity. If inventories rise too quickly, suppliers can delay expansion, reduce wafer starts, or redirect more capacity towards HBM and server memory.
Bernstein’s numbers illustrated how much room manufacturers may have. Even after its projected 2028 price drop, the firm estimates SK hynix could retain an 86.4 percent DRAM gross margin. That would make lower prices tolerable for the manufacturer without requiring it to chase maximum volume or pass every efficiency gain on to consumers.
The realistic wait is measured in years, not quarters
Late 2028 is the earliest date worth actually planning around
The genuinely good news, then, is not that cheap RAM is around the corner. It is that the first credible conditions for a broad correction are beginning to appear on the horizon. If supply expansions arrive on schedule and demand follows the more moderate forecasts, late 2028 could look significantly better than 2026 or 2027 for a major memory upgrade.
For someone who needs more memory today, however, waiting two years is hard to justify. Prices are still projected to rise in the near term, and an eventual decline at the chip level will take time to work through contracts, module manufacturers, distributors, and retailers. But anyone planning a non-essential, high-capacity build now has a plausible reason to wait. The market may finally be heading towards surplus, even if manufacturers do everything they can to make the landing gradual rather than spectacular.