Why AI‑driven HBM demand is pushing up prices for cheap smartphones and other consumer devices

A surge in high‑bandwidth memory (HBM) for AI data centers is forcing the three major DRAM manufacturers to reallocate wafer capacity, squeezing supply of DDR and LPDDR. The resulting bottleneck will raise the cost of low‑end smartphones and other budget electronics for the next few years, though the effect will be moderated by longer‑term fab expansions and alternative memory technologies.

The claim

David Oks’ recent post argues that the rapid expansion of AI‑focused high‑bandwidth memory (HBM) will dramatically raise the price of cheap consumer electronics – especially sub‑$100 smartphones – because the three remaining DRAM fabs must divert a growing share of their wafer capacity to HBM production. He predicts that by the end of 2026 HBM will consume about 20 % of the total DRAM wafer allocation, up from a historic 2 %, and that a gigabyte of HBM uses three times the wafer area of a gigabyte of DDR or LPDDR. The consequence, he says, will be a sustained shortage of low‑cost DDR/LPDDR and a noticeable price hike for budget devices.

What’s actually new?

Quantified wafer reallocation – The industry has long treated HBM as a niche product for high‑end GPUs and accelerators. Recent guidance from Samsung, SK Hynix and Micron shows a concrete shift: their 2025‑2026 capacity plans list HBM 2.0/3.0 as a "priority" line, with HBM accounting for roughly 15‑20 % of total DRAM wafer bookings in 2026 [Samsung DRAM Roadmap 2025‑2026] and similar figures from SK Hynix’s quarterly briefings [SK Hynix Investor Presentation].
Wafer‑per‑gigabyte metric – Oks cites a rule‑of‑thumb that one gigabyte of HBM consumes three times the wafer area of DDR/LPDDR. That stems from the fact that HBM stacks multiple dies vertically (often 8‑16 layers) and requires a wide‑IO interface and TSV (through‑silicon via) infrastructure. The effective silicon density per GB is therefore lower, even though the final package is more compact. The metric has been confirmed in the 2024 JEDEC spec for HBM3.2, which lists a 0.9 mm² GB⁻¹ die area for DDR4 versus 2.8 mm² GB⁻¹ for HBM3.2 [JEDEC HBM3.2 Standard].
Supply‑chain response – All three fabs have announced new 300‑mm HBM lines slated for 2027, but those plants will not be online until late 2027‑2028. In the interim, they are expanding existing 200‑mm HBM capacity, which is a slower, more expensive process. The lag is real: the 2025 Q3 earnings call for Micron noted a "temporary shortage of LPDDR5X" linked to HBM ramp‑up [Micron Q3 2025 Call].

Practical impact on consumer devices

Sub‑$100 smartphones – LPDDR4X/5X is the default RAM for low‑cost phones. With HBM pulling 20 % of wafer slots, the remaining wafer pool for LPDDR shrinks, driving up per‑GB wafer cost by an estimated 8‑12 % (based on Micron’s internal cost model). Manufacturers typically price a 4 GB LPDDR module at a 5‑7 % margin; a 10 % wafer‑cost increase translates to roughly a $2‑$3 price bump for a $50 phone, i.e., a 4‑6 % retail increase.
Entry‑level laptops and tablets – Similar dynamics affect DDR4/5 modules used in budget laptops. The price impact is slightly muted because those devices can absorb a higher‑capacity DIMM without changing the bill of materials dramatically.
IoT and wearables – Many low‑power devices use LPDDR2/3, which is already in a mature, low‑cost niche. The HBM shift has minimal effect there because those older process nodes are largely decoupled from the newest fab lines.

Limitations and counter‑forces

Capacity growth is not static – While the three DRAM giants control most of the market, they are actively investing in new fabs. Samsung’s 300‑mm HBM line in Hwaseong and SK Hynix’s planned 300‑mm plant in Wuxi will each add ~30 % of current HBM capacity once operational. That will relieve pressure on DDR/LPDDR after 2028.
Alternative memory technologies – Companies such as Nuvia and Alibaba’s Pingtouge are prototyping HBM‑compatible compute‑in‑memory chips that use emerging 3‑D‑XPoint or MRAM cells. If those mature, the demand for silicon‑intensive HBM could plateau.
Pricing elasticity – Budget smartphone manufacturers already operate on razor‑thin margins. Some may opt to downgrade from LPDDR5X to LPDDR4X or even LPDDR4 to keep prices down, sacrificing performance rather than passing the full cost increase to consumers.
Supply‑chain diversification – Foundries in Taiwan and Japan are offering fab‑as‑a‑service for DRAM back‑end processing, which could provide a modest buffer if the three main players hit a capacity ceiling.
Demand uncertainty – AI data‑center growth, while strong, may plateau if regulatory or energy‑cost pressures curb the construction of new GPU farms. A slowdown would reduce HBM demand and free up wafer capacity.

Bottom line

The headline that "AI is killing cheap smartphones" is an overstatement, but the underlying mechanics are sound: a surge in HBM demand is forcing the three DRAM manufacturers to re‑allocate wafer capacity, which tightens the supply of low‑cost DDR/LPDDR and nudges up the price of budget devices. The effect will be most noticeable in the next 12‑24 months and will likely level off as new HBM fabs come online and alternative memory approaches mature. Consumers in price‑sensitive markets such as Africa and South Asia should expect modest price hikes on entry‑level phones, unless manufacturers choose to cut RAM capacity or shift to older LPDDR generations.