A 64 GB DDR5-6000 kit, which cost $222 a year ago, is now $1,272. This is not a typo and not an isolated price from a single seller: on August 17 Tom’s Hardware published a retail price roundup for the US, and there isn’t a single line showing a rise of less than two and a half times.
On the same day TrendForce reported that the German DDR5 price index rose from 445% to 486% — that is, a typical memory kit in Germany now costs almost 4.9 times more than a year ago.
For those who run a home server or pay for VPS, this is no longer “market news somewhere.” This is about the next upgrade.
According to Tom’s Hardware, August 17, 2026, US retail prices
Numbers worth remembering
Retail in the US, August 2025 → August 2026:
| Kit | Was | Now | Growth |
|---|---|---|---|
| 16 GB DDR5-6000 | $108 | $572 | +429% |
| 32 GB DDR5-5600 | $191 | $1118 | +485% |
| 64 GB DDR5-6000 | $222 | $1272 | +473% |
| 32 GB DDR4-3200 | $222 | $614 | +177% |
| 32 GB DDR4-3600 | $300 | $789 | +163% |
A separate line — 128 GB DDR5-6400 for $3,399. That’s about ten times higher than the minimum price ever recorded for that capacity ($329).
DDR4 is rising noticeably slower than DDR5 — but “slower” here means an increase of 2.6–2.8x, not stability.
Why DDR4 lags DDR5 in growth tempo. Manufacturers are winding down older lines in favor of what data centers buy, so DDR4 grows more from supply shortages than from competition for it. The practical effect for us: DDR4 platforms remain a relatively reasonable entry point for a home server — but that window is not a constant.
Where such a jump came from
Both sources agree: the same reason, and it’s named identically — purchases for AI infrastructure. Hyperscalers contracted volumes in advance, retail gets the leftovers. Tom’s Hardware’s wording sounds almost cartoonish — mass DRAM chips today cost more than half the price of gold per kilogram.
TrendForce adds a forecasted part: contract prices for server DRAM in Q3 2026 are expected to rise 13–18% quarter-over-quarter, and the under-supply situation is expected to persist through 2027.
Forecast is a forecast. “Shortage through 2027” sounds confident, but in 2018 memory market turned down in one quarter after nearly two years of growth. Planning purchases based on “it will only get more expensive” is as risky as waiting for a drop. The only difference is that the first option costs money right now.
For China, both sources are more cautious: weekly spot price jumps for DDR5 around 14%, with a 24‑bit chip reaching $48 and a 16‑bit one up to $40. These are data from a Shenzhen marketplace quoted by TrendForce, not a direct measurement from the publication, and should be interpreted accordingly.
Scenario 1. The server is already built and running
The most common question in this situation is — should you buy memory sticks for the future? Before answering, it helps to see how much memory you actually have in use, not how much your free shows.
# how much memory can be freed without swap (available, not free)
free -h
# who is actually consuming: top-15 processes by RSS
ps -eo pid,comm,rss --sort=-rss | head -16
# how much ARC cache takes if you have ZFS
cat /proc/spl/kstat/zfs/arcstats | grep -E '^size|^c_max'
Classic home server trap: the monitor shows 90% memory usage, the owner goes to buy memory sticks — and it’s just page cache and ARC that the kernel will give back to load. Look at available in the free output, the swap amount, and whether OOM-killer has ever run (journalctl -k | grep -i "out of memory"). If OOM has never triggered and swap hasn’t been touched — nothing to buy.
Second, given current prices, stop throwing money away. Older DDR4 sticks from decommissioned machines are now real money, and home server compatibility with them is usually broader than it seems.
Scenario 2. You’re building a new server right now
Here memory may be the most expensive line item in the budget for the first time in a long time — more expensive than the CPU and the disks. What does this change:
- Budget memory first, not CPU. Decide how many gigabytes you need, then pick a platform to fit that amount.
- Don’t buy four sticks when two will do. Slots left free are an option for the future, and right now they cost more than usual.
- Look at used server platforms with DDR4 RDIMM/LRDIMM. Registered memory from decommissioned servers remains the cheapest way to get a lot of GBs — provided the motherboard supports it.
- Recalculate whether you actually need a dedicated server for every task. One node with 32 GB is cheaper than two with 16.
What not to do — buy a dual-sized kit “for future growth,” assuming prices will rise further in a year. This isn’t an upgrade; it’s a commodity-market deal, with your budget as leverage and no quick exit.
Scenario 3. You’re renting a VPS
I don’t have direct data that rental rates rose in line with hardware, and I won’t claim it without a source. What can be said with justification: the cost per GB of memory for the host increases with the market, and pricing lines are usually revised when the fleet is updated, not instantly.
From this follow two practical things. First: if you have a long-term fixed-price plan with large memory, that’s an asset now, and renewing it is cheaper than migrating for a couple of dollars. Second: if you’re planning a move to a bigger configuration, it makes sense to check the price in advance rather than when you’re hitting the limit.
The guiding principle for all three scenarios is the same: measure first, then buy. At $572 for 16 GB, the cost of a mistake like “bought with a buffer that wasn’t needed” has grown by exactly the same factor as memory.
If you’re renting a server and still choosing a configuration, you’ll find the VPS intake checklist for the first hour helpful: [link].
Has anyone run into this in practice — had to skip an upgrade or rebuild the plan? And a counter-question: how much memory is actually in use under load when you look at available, not the pretty graph in monitoring?