Why RAM Is So Expensive Right Now — And When Prices Might Come Back Down

If you've priced out a new PC, laptop, or even a game console lately, you've probably noticed something alarming: memory has become one of the most expensive components in any build. What used to be a relatively cheap, commodity part of computing has turned into a genuine budget-buster in 2026.

Just How Bad Is It?

The numbers are stark. DDR5 retail prices hit another record in August 2026, with one widely-tracked index putting prices at roughly 4.9 times what they were just over a year earlier. Even older DDR4 memory — long considered a safe, cheap fallback — hasn't been spared: a 32GB kit that cost $60–$90 back in October 2025 was going for $150–$180 by January 2026, roughly double to triple the price.

On the wholesale side, the picture is just as dramatic. Industry researcher TrendForce estimated that global DRAM contract prices rose 90–95% quarter-over-quarter in the first quarter of 2026 alone, followed by another 58–63% jump in the second quarter. Broader market analysis puts DRAM prices up 171% year-over-year, with DDR5 spot prices quadrupling since September 2025 — a rate of increase that has reportedly outpaced gold.

This isn't confined to desktop PCs, either. The shortage has pushed up the price of laptops, smartphones, and even gaming consoles — Valve's Steam Machine reportedly launched $250–$350 above its originally targeted price because of the memory crunch.

What's Actually Causing This

Unlike past memory shortages, this one isn't a simple case of manufacturers under-investing and demand catching them off guard. The root cause is AI.

Data centers training and running large AI models need enormous amounts of memory — specifically High-Bandwidth Memory (HBM), which sits directly alongside AI accelerator chips like Nvidia's GPUs. Samsung, SK Hynix, and Micron, the three companies that produce nearly all the world's DRAM, have been deliberately shifting factory capacity away from ordinary consumer DDR5 (and NAND flash) toward HBM production, because HBM is dramatically more profitable — commanding roughly three to five times the revenue per wafer compared to standard DDR5, while also consuming about three times the wafer capacity per gigabyte.

Hyperscalers and AI infrastructure projects are the driving force: capital expenditure on AI infrastructure is approaching $600 billion in 2026, up 36% year-over-year, and initiatives like OpenAI's Stargate project alone are estimated to be absorbing a huge share of available memory supply. Analysts at Goldman Sachs forecast a global DRAM shortfall of 4.9% for the year — the widest gap in 15 years — with similar shortfalls in HBM and NAND flash.

In short: manufacturers aren't struggling to keep up with demand — they're choosing where to sell their limited factory capacity, and AI customers are simply outbidding everyone else.

Why This Shortage Is Different From Past Ones

Memory prices have spiked before — notably in 2017–2018 and 2021 — but those were cyclical shortages that resolved once manufacturers ramped up production to meet demand. This time, the situation is structural rather than cyclical. Because AI demand is what's pulling capacity away from consumer memory, and that demand shows no sign of slowing, the normal "prices spike, then manufacturers build more capacity, then prices fall" cycle doesn't apply in the same way. Building new fab capacity takes three or more years, and manufacturers have little incentive to redirect existing capacity back to lower-margin consumer DDR5 while AI demand keeps HBM more profitable.

So When Will Prices Actually Go Down?

The honest answer: not soon, and possibly not back to where they used to be. Here's what the data suggests:

Short term (through late 2026): Prices are still rising, but the rate of increase is slowing. After the brutal 90%+ quarterly jumps earlier in the year, TrendForce's more recent forecasts point to "only" a 10–15% quarter-over-quarter increase in Q3 2026 for DRAM, and 13–18% for contract prices — still an increase, just a gentler one. Analysts note the market is starting to hit affordability resistance, meaning buyers simply can't absorb price hikes at the pace seen earlier in the year.

Medium term (into 2027): Most forecasts don't expect relief until at least the second half of 2026 at the earliest for the worst-hit categories like legacy DDR4, with continued pressure on DDR5 and server memory as AI demand persists. Some industry voices have suggested the supply-demand imbalance could persist well into 2027 as new fab investments take years to come online.

Longer term / possible reset (2027 and beyond): Some analysts sketch out a scenario where all the delayed fab expansion projects come online around the same time in 2027, potentially causing a sudden oversupply and price crash — sometimes described as a "bubble burst" scenario, particularly if AI investment cools or an "AI winter" reduces data center memory demand. But this is speculative, not a confirmed timeline, and depends heavily on whether AI infrastructure spending keeps accelerating.

What This Means for You

• If your current device has enough memory, there's little reason to panic-buy more at today's inflated prices — you'd likely be paying a steep premium for capacity you don't need yet.
• If you genuinely need to buy now (a failing PC, an unavoidable upgrade), expect to pay significantly more than you would have a year ago, and expect prices to keep climbing gradually rather than falling in the near term.
• DDR4 is not a safe harbor. Reduced production of older memory standards means DDR4 has seen steep price increases too, even as some buyers turn to it as a cheaper alternative to DDR5.
• This is a multi-year structural shift, not a blip. Unless AI infrastructure investment slows meaningfully, don't expect memory prices to return to 2024-era levels anytime soon. Planning around "when RAM gets cheap again" may be less useful than budgeting for memory as a persistently more expensive component going forward.

The bottom line: RAM prices are high because AI data centers are outbidding everyone else for the same factories that make your laptop's memory, and that dynamic isn't likely to reverse until either AI demand cools or a wave of new manufacturing capacity comes online — realistically sometime in 2027 at the earliest, and even then, a full return to old price levels is far from guaranteed.