The technical differences between DDR4 and DDR5 have been stable for years, and they are laid out piece by piece in our side-by-side comparison: more bandwidth, near-identical real-world latency, more capacity per stick. None of that moved in 2026.
What moved is the price. It moved so far that "which memory should I buy" has stopped being a technical question and become a budgeting one.
The numbers, with a date on them
As of August 2026, DDR5 was trading at roughly 486% of its July 2025 baseline: close to five times the price in a little over a year. DDR4, which was already end-of-life and should have been drifting down, rose between 120% and 180% depending on the module. A 64 GB DDR5 kit pushed past $1,100.
Read those figures with the date attached. They are a moment in the market, not a constant, which is exactly why this article dates them rather than presenting them as "the price of RAM".
Why it is rising: not a bottleneck, a decision
The temptation is to read this as a passing shortage of the kind that corrects itself. It is not. Manufacturers are reallocating wafer capacity away from consumer DRAM and toward HBM, the memory that AI accelerators consume, because the margin is structurally higher and the contracts run for years.
That changes the forecast. A bottleneck clears when the factory catches up; a deliberate reallocation of capacity clears when somebody builds new factories. Intel has pointed to 2028 before conditions normalise, and it is not the only voice in the industry working with that horizon.
What it means for an actual purchase
If you already own an AM4 or LGA1200 machine
Your case for staying put just got stronger. The memory already sitting in your board is the cheapest memory you will find anywhere right now: it costs nothing. Upgrading around that platform, with a second-hand processor or a better graphics card, sidesteps the one part of the budget that has exploded.
Before deciding which piece to move, the bottleneck detector tells you which one is actually holding you back. Spending on the other one is throwing money away in the worst possible year to do it.
If you are building new
There is no decision to make: every current platform from either brand is DDR5. The useful question is not "DDR4 or DDR5" but "how much DDR5". And there the advice has genuinely flipped.
The habit of buying extra just in case, 64 GB when 32 is plenty, was cheap when memory was cheap. At current prices those extra 32 GB can cost more than a whole tier of graphics card. Buy what the workload needs today and leave the slots empty.
If the work includes local language models
This is the one case where buying surplus capacity still holds up, because there memory is not a luxury but the hard limit on what you can run at all. Even so, work it out first: the VRAM calculator and the inference speed estimate usually show the constraint sitting on the graphics card, not on system RAM.
The verdict
DDR5 is still the better memory and still the only option on new platforms. What changed is that it is no longer a consequence-free purchase: every extra gigabyte now competes directly with the graphics card for the same money, and the graphics card still buys more frames per dollar than any memory generation ever has.
If your current platform works, this is an excellent year not to replace it. Check the full combination in the simulator before moving anything, and size the power supply against what you are actually going to build.
