When Will RAM Prices Come Down?

What changed and why

Unlike the hard drive market, where forecasts genuinely disagree, the memory forecasts are close to unanimous. The spread is between Q4 2027 at the most optimistic and 2028 or later at the most pessimistic, and the pessimistic end has the makers themselves on it.

Desktop memory is priced by a market it is not part of. The makers divide wafer capacity between high-bandwidth memory for AI accelerators, server DIMMs and consumer modules. HBM earns the most per wafer, server modules next, consumer least. When AI demand expanded, consumer allocation was cut first, and the price of a gaming kit became a function of how many accelerators are being built.

DRAM contract prices rose roughly 90 to 95% quarter on quarter in the first quarter of 2026, and the pressure has not eased since. TrendForce forecasts a further 10 to 15% on DRAM contract prices in Q4 2026, with AI server demand sustaining increases even as consumer demand weakens. That last detail is what breaks the usual intuition: softening consumer demand normally pulls prices down, but the consumer buyer has stopped being the marginal buyer.

Reported capacity commitments make it concrete. DRAM and HBM output for all of 2027 is described as booked and sold under long-term agreements, leaving no additional supply to release into the retail channel that year. Separately, production needs to grow around 12% a year through 2027 to meet demand against projections nearer 7.5%, which leaves makers able to fill roughly 60% of global demand by the end of 2027.

The arithmetic is what makes this unusually clear. Production needs to grow around 12% a year through 2027 to meet demand, against current projections nearer 7.5%. A gap of that shape does not close with better allocation or a demand wobble; it closes when new fabrication capacity comes online, and that takes years. Meanwhile 2027 output is already booked under long-term agreements, so even the capacity that does exist is spoken for.

The other path to lower prices is demand falling, which in this market means AI infrastructure spending slowing. That is a question about capital markets rather than about memory, and none of the sources above forecasts it. Anyone offering you a specific date for relief is going beyond what the published record supports.

Who it hits, and how hard

If you need memory now, the forecasts are not an argument for waiting. Every dated position points flat or up for at least the next year.

If you are buying speculatively, that is the purchase worth deferring, because density carries the biggest premium and you are paying it for capacity you will not use.

If you are planning a build, budget memory at current prices rather than remembered ones. A 2024 mental model of what a 32 GB kit costs will wreck a build budget.

If you are watching for a turn, watch contract pricing rather than shelf prices. TrendForce's quarterly outlooks move first; retail follows with a lag of a quarter or more.

What to do now

1. Buy what the machine needs now; defer only genuinely speculative capacity.
2. Check the live medians on our DDR5 and DDR4 pages rather than a remembered price.
3. Confirm bandwidth versus capacity with the bandwidth calculator before buying more sticks than you need.
4. Watch TrendForce quarterly contract outlooks as the earliest public signal, not retail shelf prices.
5. If you are on an end-of-life DDR4 server platform, read the registered DDR4 page: a general recovery may not reach you.
6. Work the decision through deliberately on our buy now or wait page.

Outlook

Everything above is someone else's published position with its date and a link. We publish no forecast of our own: we carry affiliate links to the prices we would be forecasting, and we have no information these sources lack.

We publish no forecast of our own. Everything above is someone else's published position, with the date it was published and a link to it, because a price forecast from a site that sells against those prices is worth nothing to you.