RAM prices have climbed sharply after manufacturers cut DRAM output and AI infrastructure absorbed high-margin memory capacity. Consumer DDR5 now competes with HBM and server RDIMMs for the same limited production resources. DDR4 is not a guaranteed escape because its production is winding down too. This guide explains what is driving the cost, how long the shortage may last and what buyers can do now.

AI memory eats more silicon

Most explanations stop at "AI demand is high." That is true, but it misses the part that matters: AI memory uses an unusually large amount of manufacturing capacity.

The fastest AI accelerators use High Bandwidth Memory, or HBM. Instead of sitting on a motherboard next to the processor, HBM stacks several memory dies and places them close to the accelerator. That gives GPUs the bandwidth needed to train and run large models, but it also makes HBM expensive to produce.

During its fiscal second-quarter 2025 earnings call, Micron said HBM3E requires about three times as much silicon as DDR5 to produce the same number of bits. It expects HBM4E to push that ratio beyond four to one. Every gigabyte of HBM can therefore displace several gigabytes of ordinary DRAM production.

Industry estimates put HBM at about 22% of the three largest manufacturers' DRAM wafer input by the end of 2026 while accounting for only about 9% of the DRAM bits they ship. On September 29, a Samsung executive said HBM already used about 20% of industry wafer capacity and could approach 30% in 2027.

That is the first squeeze on desktop RAM: a growing share of the silicon goes into a product that yields fewer bits per wafer.

Ordinary server RAM is taking the rest

HBM gets the headlines, but cloud providers also need conventional system memory. An AI server still needs DDR5 RDIMMs for the operating system, model data, caches and all the work that happens around its accelerators. As companies move from training models to serving them to millions of users, that demand keeps growing.

Memory manufacturers have responded by prioritizing high-capacity server modules. Suppliers are sending additional output to server RDIMMs before consumer products. By early 2026, a wafer turned into 64GB DDR5 server modules could earn more than one used for HBM.

The same factories make server and consumer DRAM. They cannot maximize both at once. When a cloud provider offers a long contract for higher-margin server memory, a $100 gaming kit is an easy product to push down the schedule.

Manufacturers cut supply before demand exploded

The timing made the shortage worse. Memory prices crashed in 2022 and early 2023 after the pandemic hardware boom ended. Manufacturers had too much inventory, so they pulled back.

Samsung announced a "meaningful" production cut in April 2023. SK hynix and Micron reduced output too. Those cuts made sense while warehouses were full and prices were falling. Then generative AI took off, data-center orders surged and the industry had little spare capacity ready to absorb them.

Global DRAM supply is expected to grow by only 16% in 2026, below the industry's historical rate. That sounds like a lot until demand is growing faster and every new HBM bit uses several times the silicon of a DDR5 bit.

Restarting old lines helps at the edges. It does not create a new fab.

New memory factories will not fix prices soon

A semiconductor fab is one of the slowest and most expensive factories to build. After construction, the equipment must be installed, calibrated and qualified. The first usable wafers arrive years after the original announcement, and full production comes later still.

Micron's fiscal third-quarter 2026 filing gives a useful timeline. The company expects first wafer output from its new Idaho fab in mid-2027. A second Idaho fab is scheduled for late 2028. Its planned New York site will not contribute until 2030 or later.

Those projects add capacity eventually, but they do nothing for a PC build this month. Even when new fabs open, there is no guarantee that their output will go to desktop DIMMs. If HBM and server contracts remain more profitable, those products will keep winning the allocation fight.

IDC describes this as a potentially permanent reallocation of global wafer capacity rather than a normal memory cycle. "Permanent" may prove too strong, but the old pattern of adding supply until consumer prices collapse is no longer the only possible outcome.

PC builders are last in line

Large cloud companies buy memory differently from consumers. They negotiate directly with manufacturers, order enormous volumes and sign contracts months or years ahead. By January, cloud providers were already locking in capacity and leaving other buyers to accept higher prices.

Smartphone and PC makers at least have enough scale to negotiate. DIY builders do not. Do-it-yourself systems, which often means gaming PCs, are expected to carry the heaviest burden. Retail RAM sits at the end of the chain, after AI accelerator makers, cloud providers and major device manufacturers have secured supply.

Micron's decision to close its Crucial consumer business made that shift unusually visible. The company stopped shipping Crucial products to retailers in February 2026 so it could focus on larger strategic customers. Crucial kits still listed at stores are likely remaining inventory rather than fresh production.

One of the best-known consumer RAM brands did not leave because people stopped buying RAM. It left because selling to consumers was no longer the best use of Micron's output.

Prices rose much faster than production costs

Scarcity explains why prices can rise. It does not mean a RAM kit suddenly costs five times as much to manufacture.

Micron's fiscal third-quarter 2026 filing shows DRAM sales rising 211% over nine months, driven mostly by an increase of roughly 140% in average selling prices. Quarterly revenue grew about 4.5 times year over year while production costs rose by around 10%.

The margin change is even clearer. The same business unit went from a 24% gross margin to 87%. Supply is tight, buyers are competing for capacity and manufacturers can charge prices that would have looked absurd a year earlier.

Calling this pure price gouging would be too simple. The shortage is real, HBM does consume more silicon and new fabs do take years. But today's retail prices reflect pricing power as much as higher production expense. The memory companies are not merely passing on costs.

How long will expensive RAM last?

Nobody has a reliable end date.

SK hynix CEO Kwak Noh-Jung expects 2027 to be the worst year of the shortage and says demand could exceed supply beyond 2030. That view fits the fab schedules and the long contracts already signed by cloud companies.

Acer CEO Jason Chen takes the opposite position. He calls a shortage lasting to 2030 impossible, expects prices to fall after the middle of 2027 and accuses memory manufacturers of talking up the shortage to protect their margins.

Recent pricing supports both arguments a little. Consumer resistance slowed third-quarter increases to 13–18%. Buyers do have a limit. The fourth-quarter outlook then rose again because cloud demand stayed strong. Prices can slow without returning to normal.

The safest conclusion is less satisfying: relief before mid-2027 looks unlikely, and a return to 2025 prices is not guaranteed. A global recession, an AI spending pullback or faster capacity expansion could change that. None is something a PC builder can plan around.

Should you buy RAM now or wait?

If your current PC has enough memory, wait. A working 32GB system does not need a 64GB upgrade just because 64GB would be nice. At today's prices, buying capacity you may use later is an expensive bet.

If a lack of RAM is already slowing your work, waiting indefinitely also has a cost. Buy the capacity you need, skip premium speed bins and compare stores before checking out. For most DDR5 desktop builds, a decent DDR5-6000 kit is a better target than a 7600 or 8000 MT/s kit whose extra speed will barely show outside a benchmark.

Use the RAM-Mageddon DDR5 comparison and sort by price per gigabyte. Retailers are receiving stock at different times, so the gap between two otherwise similar kits can be large.

A few rules make the decision easier:

• Keep a machine that already meets your needs. This is a bad market for a cosmetic upgrade.
• Buy enough capacity in one matched kit. Adding a separately purchased stick later can cause compatibility problems, and the second stick may cost more.
• Pay for capacity before speed. 32GB of ordinary DDR5 is usually more useful than 16GB with an impressive frequency printed on the box.
• Check the whole platform price. A prebuilt PC or laptop bought under an older supply contract may cost less than assembling the same specification from retail parts.
• Do not assume DDR4 is the escape route. DDR4 production is winding down, so its price can rise even while the technology gets older.

Not sure whether you need 16GB, 32GB or 64GB? Read our 2026 RAM capacity guide before spending anything. If you are choosing a platform, our DDR4 vs DDR5 comparison explains where each generation still makes sense.

FAQ

Why did RAM prices suddenly get so high?

The increase looks sudden at retail, but its causes have been building since 2023. Manufacturers cut DRAM production after a market crash, then AI companies began buying huge volumes of HBM and server DDR5. Supply could not respond quickly, so consumer memory prices climbed.

Is AI really responsible for expensive desktop RAM?

Yes, though AI companies are not buying the same desktop kits. HBM, server RDIMMs and consumer DDR5 compete for wafer capacity. HBM also needs about three times as much silicon as DDR5 for the same bit capacity, so its growth removes a disproportionate amount of potential consumer supply.

Will RAM prices go down in 2027?

They may soften after mid-2027, but forecasts disagree. New fab output should start helping, while cloud demand and long-term contracts could keep supply tight. Do not count on a quick return to 2025 prices.

Is DDR4 cheaper than DDR5 now?

Sometimes, but not consistently. Manufacturers are retiring DDR4 production, which can create shortages and price spikes of its own. Compare the total cost of the CPU, motherboard and memory rather than choosing a platform from RAM prices alone.

Should I settle for 16GB because RAM is expensive?

Only if 16GB fits your workload and budget. It remains usable for office work and lighter games, but 32GB is the more comfortable target for a new gaming or development PC. Buying too little and replacing the kit later can cost more than buying the right capacity once.

Sources

• RAM-Mageddon price comparison (prices as of September 30, 2026)
• TechFuelHQ, DDR5 RAM Buying Guide (September 2026)
• IDC, Global Memory Shortage Crisis (updated August 2026)
• Micron, FQ2 2025 earnings call transcript
• TrendForce, HBM share of wafer input (June 2, 2026)
• TrendForce, Samsung sees HBM taking nearly 30% of DRAM capacity (September 29, 2026)
• TrendForce, First-quarter contract prices and server RDIMM priority (June 1, 2026)
• TrendForce, Cloud providers lock in memory capacity (January 5, 2026)
• TrendForce, Third-quarter memory prices (July 3, 2026)
• TrendForce, Fourth-quarter memory price forecast (September 2026)
• AFP via RTHK, Slump in memory chip sales hurts Samsung Electronics (April 2023)
• AFP via Prothom Alo, SK hynix and Micron production cuts
• Micron, Form 10-Q for FQ3 2026
• Micron, FQ3 2026 results
• Micron, Exit from Crucial consumer business
• Reuters via Gulf Times, SK hynix CEO sees worst memory shortage in 2027
• Notebookcheck, Acer CEO disputes a shortage lasting to 2030