Why RAM and SSD Prices Tripled in 2026 (and What to Do)

By LK Wood IV · 2026-06-28 · ~11 min read · St. Louis County, MO

If you priced a PC build in 2024 and went back to your cart in 2026, the shock isn’t the graphics card. It’s the two parts nobody used to think about: the RAM and the SSD. They quietly became the most expensive surprise in a modern build, and the reason has almost nothing to do with PCs. Hard drives are the third casualty, on a slower fuse; that half of the story has its own page.

This is the explainer: the real numbers (cited, not vibes), the actual cause, the honest timeline, and a practical plan for building or upgrading during the crunch.

The numbers: how bad it actually got

RAM. On June 3, 2026, Tom’s Hardware’s daily RAM price tracker recorded the cheapest in-stock 32GB DDR5 kit in the US at $374.97 — a kit class that routinely sold for $80-120 a year earlier. High-capacity 64GB kits that sat near $150-200 for most of 2025 now commonly list at $600 or more. Across the board, consumer DRAM rose roughly 300-600% off its 2024-2025 lows, and DRAM contract prices jumped 55-60% in Q1 2026 alone.

SSDs. Storage followed, with a lag and then a vengeance. NAND flash pricing more than doubled in roughly six months, and by mid-2026 NAND had risen close to 4x off its lows. A 2TB NVMe drive that could be found for $120-150 a year ago now lists at $300-480; Samsung’s 990 Pro 2TB bottomed near $189 in 2025, while the newer 9100 Pro 2TB was listed around $478 in April 2026. Consumer SSDs broadly climbed 100-200% in a matter of months.

To put it plainly: in a typical mainstream build, the memory and the boot drive went from “rounding error” to the two parts that can cost more than the CPU.

Why is RAM so expensive? Your DIMM lost a bidding war to a server

The shortage is not a manufacturing breakdown. It’s a capacity reallocation, and the buyer who outbid you is an AI data center.

Memory makers have a finite amount of wafer capacity, and in 2026 they pointed it at the most profitable product they make: high-bandwidth memory (HBM) and high-capacity modules for AI accelerators. IDC and other analysts describe a deliberate shift of production away from consumer electronics toward AI memory, with AI workloads projected to absorb about 20% of global DRAM wafer capacity in 2026.

The economics are brutal and they explain everything:

• A single HBM3E stack sells for roughly $60-100, versus about $5-10 for a comparable amount of conventional DDR5.
• One gigabyte of HBM consumes roughly the wafer area of four gigabytes of standard DRAM.

So a wafer turned into HBM earns several times more than the same wafer turned into the DIMMs you buy — while also producing less total memory. Every gigabyte of AI memory is effectively several gigabytes of consumer memory that never gets made, at many times the margin. Faced with that math, no manufacturer rationally prioritizes a $90 desktop kit. NAND tells the same story for storage: AI data centers need vast, fast storage, and one major SSD-controller maker (Phison) confirmed NAND prices had more than doubled and that all of 2026’s production was already sold out.

That’s the uncomfortable part: this isn’t a glitch that gets patched next quarter. It’s the market working exactly as designed, against you.

When does it end?

Honestly: not soon, and not on a schedule anyone can promise. But the cited consensus is consistent enough to plan around.

• Through 2026: prices stay elevated and keep climbing, with the peak expected in Q3-Q4 2026. TrendForce projected client SSD contract prices rising at least 40% quarter-over-quarter in Q1 2026, with tight supply persisting all year.
• Relief keeps moving out. The June consensus said late 2026. TrendForce’s July 30 outlook now has DRAM undersupplied by 1-2% for all of 2026 and the gap widening through 2027, and the analyst read in PCWorld’s investigation puts the start of real improvement at Q4 2027, with normal pricing not expected before 2030.
• Memory and storage now diverge. The same TrendForce release expects NAND supply to turn positive in 2027 while DRAM stays short. SSDs likely come off the boil first. RAM does not follow on the same clock.
• Real capacity: new fabrication capacity is generally expected to come online around 2027-2028 — and even that only helps if AI memory demand moderates, which few analysts expect.

The blunt takeaway from the timeline: in 2026, “wait for prices to drop” is bad advice, and the July forecasts only hardened that. Prices are far more likely to be higher in three months than lower.

What to actually do (a builder’s plan)

Here’s the practical part, and it’s where most of the value is.

1. If you need RAM or an SSD, buy it now. This is the rare case where “buy the dip” is wrong — there is no dip coming in 2026. If a build or upgrade is already on your list, the memory and storage are the two parts to lock in first. As of August 2026 that is no longer a head start on the Q3-Q4 peak – we are in it.

2. Right-size, don’t over-buy. The instinct to “future-proof” with 64GB or a 4TB SSD is exactly the wrong move at these prices. For most people:

• RAM: 32GB DDR5-6000 is the sweet spot for gaming and general use; 16GB is the floor if the budget is brutal (Tom’s index still had 16GB DDR5 at $214-239 on its August 24 update). Going to 64GB roughly doubles an already-painful cost for capacity most desktops never touch. (How much RAM do you actually need in 2026? breaks it down by use case.)
• SSD: a single 1-2TB NVMe covers the vast majority of builds. Buy the capacity you’ll use in the next year, not the next five.

3. Reuse what you already own. DDR5 and NVMe carry forward cleanly. If you’re upgrading a platform, pull the RAM and the boot drive from the old machine before you price new ones — a perfectly good DDR5 kit or 1TB NVMe you already paid 2024 prices for is the cheapest memory you will buy in 2026. The flip side applies too: if a machine is blue-screening with MEMORY_MANAGEMENT and you suspect a stick, run the free tests before replacing anything — at these prices a misread memory error is a $470 mistake (the cheapest qualifying 32GB kit in the price index, August 24 capture).

4. Spend the budget where the spike isn’t. GPUs, CPUs, and motherboards have been comparatively stable through the memory crunch. That means the RAM and SSD are the line items distorting your build budget — so cut elsewhere only if you must, and treat memory/storage as the fixed cost they’ve become. Our $1,000-class budget AM5 build was priced around this shortage and shows how much the memory line alone moves the total.

5. Model the running cost, not just the sticker. If you’re building a homelab or always-on box, the parts you keep for years cost more than their purchase price. Size the rest of the system honestly and run the numbers with the Power & Cost Calculator so the shortage doesn’t push you into an oversized, power-hungry box you’ll pay for monthly.

Those five moves assume the purchase is happening. The three questions below decide whether it should.

Should you wait for RAM prices to drop?

Only if you can wait years. Not months. TrendForce’s July 30 outlook has DRAM undersupplied through 2026 and the gap widening in 2027, and the analyst read in PCWorld’s investigation does not call pricing normal again before 2030. Against that forecast, “wait for the dip” is not a strategy. It is a decision not to build, which is a fine decision if the machine you have still works. Keeping a working box in service through the crunch is the one move that costs nothing.

If the build is genuinely needed, buy the memory line first and work the bundles. When I priced the combo route at Newegg on August 25, CPU-plus-board-plus-RAM bundles were taking $70-104 off the summed list prices, which is the closest thing to a discount the DRAM market offers right now. And if the itch is an upgrade rather than a build, put the money in the part that didn’t spike. A GPU swap into an existing box sidesteps the memory market entirely.

What DDR5 costs right now runs this wait-or-buy call against each capacity tier in more depth.

The DDR4/AM4 escape hatch

There is one, I priced it on August 25, and it is narrower than the folklore says.

The memory half holds up. On Tom’s Hardware’s August 24 index update, the cheapest in-stock 32GB DDR4-3200 kit was $179 against $402 for 32GB of DDR5-6000. Call it $220 saved on the memory line, real money, from the cheap half of a market that did not stay cheap anywhere. DDR4 doubled to tripled too; the same index had those kits at $60-90 in October 2025.

The CPU half has a story. Used Ryzen 7 5800X3D prices went vertical when DDR5 spiked; Tom’s Hardware tracked eBay averages of $500-600 last December, above what a new 9800X3D cost. AMD’s answer was to restart production. The 5800X3D 10th Anniversary Edition went back on sale June 25 at $349, and AMD told Tom’s at Computex it is an ongoing product rather than a limited run. It was in stock at Newegg, sold by Newegg, at exactly $349 when I checked on August 25. A bundle pairing it with a B550 board and 16GB of DDR4-3200 ran $519.99 the same day. Do not pay $500 for a used one now. The reissue reset that market.

Who the hatch is for. If you already run AM4, you are holding the best hand in this whole mess. A 32GB DDR4 kit at $179 is the last reasonably priced meaningful upgrade in the DIY market, and your platform cost is zero. Board makers restarted DDR4 motherboard production into 2027 precisely because of this demand, so parts will be around. For a from-scratch build the math is tighter. You save roughly $220 on RAM and often $100 or more on the board, and in exchange you accept a socket whose CPU line ended with Zen 3 and kits that now top out near DDR4-3600 because the fast dies are out of production. Defensible for a budget gaming box that has to exist this year. Wrong for a build you intend to grow.

I have not run DDR4 through our own price index; it tracks DDR5-6000 kits only, so the DDR4 figures above lean on Tom’s tracker rather than on first-party captures.

16GB or 32GB under duress

From the same August 24 table. 16GB of DDR5-6000 was $239. 32GB was $402. That is 68% more money for twice the memory, so the 32GB kit is still the better per-gigabyte buy, and it stays the right call whenever the budget can stretch. What changed is the fallback. 32GB of DDR4-3200 at $179 now costs less than 16GB of DDR5 — if the real choice is a 16GB AM5 build or a 32GB AM4 build on the same memory budget, capacity argues for the older platform.

Hard-capped on AM5 anyway? 16GB still handles 1080p gaming and general desktop work today. Know what you are signing up for. It is the first ceiling you will hit, and “buy 16GB now, add more later” is a bet that prices fall while every forecast above says they will not. How much RAM you need sorts capacity by workload rather than by price.

For the deeper RAM-specific breakdown — speeds, timings, and what actually matters when you do buy — see the DDR5 RAM buying guide, which now carries a dated note reflecting these 2026 prices.

The bottom line

The 2026 memory crunch is real, it’s cited, and it’s structural: your DDR5 kit and your NVMe drive lost a bidding war to AI data centers that pay several times more for the same silicon. Prices were still climbing through mid-2026, and the July forecasts pushed DRAM relief past 2027. Nobody credible expects 2024 levels back soon.

So the move isn’t to panic and it isn’t to wait. It’s to buy what you need now, buy exactly as much as you’ll use, reuse what you own, and put the rest of the budget where the spike isn’t. If the DDR5 math still refuses to close, the DDR4/AM4 hatch is open and honestly priced. Do that, and the most expensive surprise in a 2026 build becomes a line item you planned for instead of one that blows up your cart.