Analysis: DRAM and NAND memory prices have jumped 600%+ over the past year for routers and set-top boxes, hitting telcos targeting aggressive broadband rollouts
MS Hwang is a research director at Counterpoint, specializing in memory semiconductor research.
Counterpoint Research
Context & Ripple Effects
This is an escalation of a memory-cost problem that was already visible when DRAM prices doubled between May and June 2025. The impact now reaches broadband access equipment, where routers and set-top boxes are part of operators’ rollout economics.
Related coverage shows the pressure is not confined to network gear: later reporting tracked sharply higher contract memory prices and memory-driven premiumization in lower-cost smartphones. That makes this a cross-device component-cost issue rather than an isolated telecom procurement event.
First-order effects
Telcos pursuing broadband expansion face higher unit costs for memory-equipped routers and set-top boxes, tightening the budgets allocated to those deployments.
Equipment vendors supplying those devices must absorb, pass through, or renegotiate around a substantially more expensive DRAM and NAND bill of materials.
Second-order effects
Operators and vendors are likely to reassess device specifications, procurement timing, and subsidy levels as memory costs make aggressive rollout targets more expensive to meet.
The same price shock across consumer electronics and access hardware increases competition for cost-sensitive memory supply, as reflected in the subsequent rise in NAND contract prices.
Third-order effects
If sustained, memory pricing becomes a constraint on broadband adoption economics as well as on device affordability, shifting attention from headline network investment to the cost of the endpoint hardware required to use it.
The pattern supports a more cyclical hardware market in which capacity and component pricing can determine which product tiers and infrastructure rollouts remain commercially viable.
The trend: This is part of a memory supercycle in which DRAM and NAND cost inflation is propagating from chips into consumer devices and telecom access equipment.
Something isn't right here. What rack servers would utilize mechanical hard disks for AI? Perhaps if for storage of the LLMs, but makes no sense for anything else. Mechanical drives have absolutely poor performance when compared to solid state. But again, cold/long term/bulk
Western Digital's HDD production capacity for 2026 is fully sold out, CEO Irving Tan confirmed in recent statements, due to AI demand. Firm orders from top customers cover the entire year, with long-term agreements extending into 2027 and 2028 for some. [image]
Western Digital has no HDD capacity left, as its CEO reveals massive AI-driven deals have consumed available supply. 🔗 https://wccftech.com/... [image]
Many of data centers' externalities are not unique to their use for AI; they all use a ton of electricity, are eyesores, employ relatively few people given their economic impact, etc., whether they're serving Claude or serving Netflix. But supply strain is new and overwhelmingly…
I haven't seen most non-tech reporter critics of AI talk about this effect it's having immediately on the economy: the prices of memory, storage, and every object with any of these in it are going through the roof as data centers gobble them up in bulk [embedded post]
Western Digital has recently announced that it sold all of the year's hard-drive inventory to AI companies. — In other words, prepare for huge SSD price surges. — https://wccftech.com/...