The server memory shortage remains in force. The live medians below are the market's current state — computed from 49 in-stock modules (29 DDR4, 20 DDR5) on the same refresh cycle as every price on this site. Of the modules with a valid week-over-week comparison, 0 rose and 0 fell. Full figures, the weekly methodology and the trend chart live on the price index.
No week-over-week moves above the noise floor in the tracked basket — flat weeks happen; the index above is the level to watch.
Moves are computed from each module's daily price history over the trailing 7 days; modules without history yet (recently added) do not appear until they accrue it.
Everything on this page traces to the structural analysis in our memory-shortage report, which remains the reference: (1) Samsung, SK Hynix and Micron — over 95% of global DRAM output — have shifted wafer capacity to High Bandwidth Memory for AI accelerators, which consumed roughly 23% of DRAM wafers in 2026 versus 8% in 2024. (2) HBM earns three to five times the revenue per wafer of conventional DDR5, so the reallocation is deliberate and durable. (3) SK Hynix has confirmed its entire 2026 HBM production is already committed. (4) Micron's CEO has pointed to supply tightness continuing into 2027. (5) The measured price result: DDR4 ECC up roughly 60-80% and DDR5 ECC up 100-116% between early 2025 and Q1 2026.
Nothing in the recent data changes that picture: this is a manufactured scarcity that persists as long as HBM margins stay attractive, with the most likely path being gradual stabilization in late 2026 into 2027 rather than a return to early-2025 pricing. When those facts change, this page and the underlying report will change with them.
The framework we apply to drives on the price forecast applies to memory with one sharpening: the supply-side facts above argue against waiting for relief that analysts do not expect before 2027. If a server needs memory now, buy the requirement now — the risk of waiting is priced in the index's direction of travel, not in your favor. If the need is elective, DDR4 is the value refuge: its live index sits far below DDR5's and its platforms are mature, which is why our DDR4 tracker is where budget builds should start. For DDR5 platforms there is no substitute — track the specific module's 90-day chart on its detail page and buy on its dips rather than waiting for the market's.
One tactical note this tracker keeps proving: individual modules move more than the index. A single listing dropping 10% while the average sits flat is common — which is exactly what the "what changed" list above surfaces, and why per-module history beats market timing for most buyers.
The compressed timeline, from our reporting: through 2024, HBM's share of DRAM wafers roughly matched a niche product's — about 8%. The AI accelerator buildout changed the arithmetic in 2025: with HBM earning three to five times the revenue per wafer of conventional DDR5, Samsung, SK Hynix and Micron reallocated capacity deliberately and publicly. By October 2025, SK Hynix reported its entire 2026 HBM production already committed. Through late 2025 and into 2026, conventional DRAM supply tightened exactly as Sapphire Rapids and EPYC Genoa fleets ramped DDR5 demand — and by Q1 2026 the measured damage stood at roughly 60-80% increases for DDR4 ECC and 100-116% for DDR5 ECC against early-2025 baselines, with HBM consuming roughly 23% of global DRAM wafers.
What distinguishes this episode from the DRAM cycles of 2017-18 or 2021 is intent: previous shortages were demand surprises against planned capacity, resolved by fabs catching up. This one is a portfolio decision by an industry with three meaningful suppliers, sustained precisely because it is profitable. That is why the analyst consensus we track points to gradual stabilization rather than a crash — the suppliers are not racing to fix a mistake; they are managing a margin opportunity — and why this tracker watches supplier commitments as closely as prices. The facts above change when earnings calls change them, and this page will say so the week that happens.
Four signals drive updates here. First, the index level itself — the DDR4 and DDR5 averages at the top, recomputed on every price refresh. Second, per-module moves from daily history, because listing-level drops precede market-level ones and are the buyer's actual opportunity. Third, basket depth: modules entering or leaving stock is itself shortage data — thinning stock at rising prices reads very differently from thinning stock at falling ones. Fourth, the supply-side record: manufacturer commitments, wafer-allocation reporting, and platform-demand milestones from our news desk. When any of the four shifts materially, the status strip above changes — which is the entire point of a rolling status page over a quarterly report.
A note on what a "flat week" means in a shortage: it is not nothing. Elevated-and-stable is the shortage working as the suppliers intend — prices holding at levels that would have been unthinkable two years ago, with neither the relief of decline nor the drama of further spikes. Readers checking this page weekly should calibrate accordingly: the newsworthy events are breaks in the plateau, in either direction, and the per-module list above is where they show up first.
Running DDR4 today: you hold the shortage's best hand. Max out the platform you own at DDR4's comparatively sane prices, buy spares with the fleet, and defer the DDR5 migration until either the premium compresses or the platform genuinely ages out. The cross-generation gap on the price index is your standing argument in any budget meeting. Forced onto DDR5 by a new platform: buy complete populations in single orders, watch specific part numbers rather than the market, and treat listing-level dips — surfaced in the moves list above — as your only real discount mechanism. Waiting for the market itself has lost all cycle.
Homelab and workstation buyers: the refurbished DDR4 market is your shortage exemption — tested pulls with no wear mechanism at prices the DDR5 world can only envy, detailed on the DDR4 tracker. Procurement planning ahead: assume today's levels through at least mid-2027 in budgets, treat any analyst-projected relief as upside rather than plan, and revisit this page's status strip before each quarterly commit — it exists so that the week the facts change, your plans can.
A status page earns trust by stating its falsifiers. Three developments would move this tracker off "shortage in force": a sustained multi-week decline in the DDR5 index average (not one listing, the basket), a manufacturer publicly reallocating wafer capacity back toward conventional DRAM — the reverse of the HBM shift that caused all this — or contract-market reporting showing server DRAM prices breaking downward ahead of retail. Conversely, the signals that would darken the outlook further: HBM commitments extending deeper into 2027, DDR4 end-of-life announcements accelerating the sunset of the value refuge, or basket depth thinning at rising prices. We watch all six; the status strip above is where the first confirmed change lands, with the supporting facts updated in the same edit.
One meta-observation this tracker has earned the right to make: during a shortage, stale information is not merely unhelpful — it is expensive, because every figure people remember is lower than the one they will pay. That is why this page binds its freshness signals to the actual data refresh rather than a publish date, and why the honest answer to "what does server RAM cost" is always a timestamped one.
Check it before any memory purchase and before each planning cycle; the status strip and the moves list are deliberately the first things on the page because they are the two things that change. Everything below them is context that changes slowly and gets edited when it does — this page carries the same four freshness signals as every live page on this site, so the timestamp beside "last checked" is the honest age of what you are reading. When the strip's verdict finally changes, the buyers who were watching will have weeks of advantage over the ones who were remembering — which is, in one sentence, the entire case for a rolling status page over a news article that ages.