Build2 distinct publishers3 min readUpdated
Tom's Hardware tracks DDR5 up near 500% in twelve months, and hyperscale buyers have reportedly pre-bought almost all of 2027's global DRAM output. Deferring the purchase is not a plan.
The Engineer · Build desk

Compiled by The EngineerSomething wrong?How this is made
Tom's Hardware reported in August 2026 that 128GB DDR5 kits are selling for $3,399, ten times the lowest price the site has ever tracked, with year-over-year increases nearing 500% on high-capacity kits [1][2][3]. If your 2026 capacity plan carries memory as a boring, predictable line item, that plan is void, and the usual response of pushing the buy a few quarters right does not work here: hyperscale buyers have reportedly already locked in almost all global DRAM production capacity for 2027, with advance deposits [4].
The mid-range is where fleet refresh actually lives, and the mid-range moved too. A 64GB (2x32GB) DDR5-5600 kit that was under $200 last summer is now over $1,100 [5], more than a 5x move [2]. A kit that cost $72 a year ago is $392 [6], roughly 5.4x [1]. The comparison is PCPartPicker average retail prices for August 2026 against a year earlier, and the author calls the data somewhat approximate [7], so treat it as a directional index for consumer kits rather than as your quote.
Downgrading platforms is not an exit either. With DDR5 out of reach for most builders, the scramble for older DDR4 boards pushed DDR4 kits up 120% to nearly 180%, including one kit that went from $105 to $281 [8][9], about 2.7x [5]. In Europe, ComputerBase reports average RAM prices up 345% against September 2025, with hard drive and SSD prices both up over 125% in the same window [10][11]. That last figure matters if you were planning to buy your way out of a memory constraint with storage tiering.
Run the arithmetic on your own sizing sheet before the next planning meeting. At $3,399 for 128GB, retail DDR5 is about $26.55 per gigabyte [3]. A single 512GB CPU-memory host is therefore around $13,600 in DRAM alone at that rate [4], against roughly $2,700 for the same arithmetic a year ago on the same tracker [6]. That is the actual cost of the self-hosted-inference thesis right now, and it lands in the same week that Latent Space's AINews recap described Moore's Law as reversed for memory while celebrating a 27B open model with 262K context as locally runnable frontier-ish [16][17].
The supply side has no reason to accommodate you. PC and smartphone makers are described as fighting over scraps, ranked as low-priority buyers next to AI datacenter buildouts [13], and all four memory vendors, SK hynix, Samsung, Micron and CXMT, have doubled, tripled or better their revenues inside a year [14]. Mainstream DRAM chips are now worth over half as much per kilogram as gold [15].
Watch the vendor guidance, because it is unusually explicit. SK hynix CEO Kwak Noh-jung has warned that 2027 will be the worst year for memory supply in the industry's history, with demand outstripping production well into 2030 [18]. ADATA chairman Simon Chen put the crisis at possibly another ten years [19]. Tom's Hardware' own read is that consumer relief requires a major contraction in the AI market [20], which is to say the two things a buyer might hope for are mutually exclusive. Re-price memory and storage in every 2027 model you own, and get the quantity commitments signed rather than optioned.
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Memory prices have climbed 500% in 12 months, with year-over-year increases nearing 500% for high-capacity DDR5 kits.
128GB of DDR5 now costs $3,399.
128GB DDR5 kits are fully ten times more expensive than the lowest price Tom's Hardware has ever tracked.
Hyperscale buyers have reportedly already locked in almost all of the global DRAM production capacity for 2027, handing over advance deposits to guarantee supply.
A standard 64GB (2x32GB) DDR5-5600 kit that would have cost under $200 last summer is now demanding over $1,100.
A memory kit that was $72 last year now costs $392.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Retail price series are concrete and cross-checked; supply-side claims are second-hand
The pricing core rests on named, dated series — Tom's Hardware's own RAM price tracking plus PCPartPicker averages for August 2026 versus August 2025 — and is independently echoed by ComputerBase's European figures, which is real corroborating measurement. It is weakened by the author's own caveat that the data is approximate, by the fact that Latent Space adds no independent data (it quotes Tom's Hardware verbatim), and by the load-bearing 2027 capacity lock-in being reported with no named source or contract detail.
Price shock is already realised across regions, memory generations and adjacent components
This is not an announced plan awaiting uptake: the increases are observed in transacted retail averages across DDR5 and DDR4, in the US and Europe, and are spilling into HDDs and SSDs, alongside a visible behavioural response (builders retreating to AM4/LGA1700, which itself lifted DDR4 pricing). Vendor revenue multiples indicate the demand shift is being captured commercially. Scope limit: everything measured here is retail/consumer channel; enterprise contract pricing and server DIMM availability are not evidenced.
Numbers hold up; the apocalyptic framing and decade-long forecasts run ahead of them
The near-term price claims are close to aligned with the evidence, so the gap is modest. It is positive rather than zero because the rhetoric ('RAMageddon', 'divorced from reality', Moore's Law reversed to 2007 levels) and the forward claims outrun what is shown: a ten-year DRAM crisis and demand outstripping supply into 2030 are vendor-sourced forecasts, the 2027 capacity lock-in is unattributed, and the '10x' figure is measured against an all-time low rather than a normal baseline, which flatters the headline. Latent Space amplifies the framing without adding evidence.
Scarcity narrative benefits the quoted suppliers; publisher benefits from deal traffic
Both forward-looking pessimists quoted have commercial exposure to the shortage narrative: SK hynix's CEO and ADATA's chairman both sell into a market where tight supply supports pricing power, and the ADATA chairman explicitly dismisses an AI bubble bursting. On the publishing side, Tom's Hardware routes readers twice to its own RAM price-tracking and deals post and solicits newsletter and preferred-source signups, so urgency has an audience-monetisation value. Latent Space's incentive is aggregation reach rather than commerce. No sponsorship or paid placement is disclosed in either source, so this is structural incentive, not evidence of distortion.
High confidence in the price move, lower in the structural and forward claims
Two published sources, both dated within days, with disclosed methodology and one independent regional cross-check give solid confidence that consumer memory prices have risen roughly as described. Confidence is capped below the high band because the cluster is effectively single-origin (Latent Space quotes Tom's Hardware), the pivotal 2027 capacity claim is unattributed, the derived per-host figures are arithmetic extrapolation from retail kit prices rather than observed server BOMs, and enterprise contract pricing is absent.
build
China's accelerator swap makes Cambricon supply, not export policy, your ship-date risk1 distinct publisher
invest
A trillion-dollar memory stock is now Boise's biggest concentration risk1 distinct publisher
invest
Binance lists CXMT perps days after an $8.6B IPO, and the venue line blurs further1 distinct publisher
invest
Korean mortgage renewals reset 0.5 to 0.8 points higher, and the ETF desks smell an opening1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 19, 2026
1 article · August 17, 2026