Memory Makers Shift Production to AI, Leaving Consumer Markets Short

A Market Reorientation with Lasting Consequences

Dynamic random access memory prices have climbed 400 per cent since 2024, according to JP Morgan data, as the three companies that control roughly 90 per cent of global DRAM production - Micron Technologies, Samsung Electronics, and SK Hynix - redirect manufacturing capacity toward high-bandwidth memory chips used in artificial intelligence infrastructure. The shift has created cascading shortages across consumer electronics, with Nvidia's GeForce RTX 5090 now retailing nearly $5,000 above its January 2025 launch price of $1,999.

The supply strain reflects a structural change in how memory producers allocate limited wafer fabrication capacity. High-bandwidth memory, which AI developers require for both training and inference workloads, commands profit margins roughly 30 per cent higher than standard DRAM. Samsung executives indicated in September that HBM could represent nearly 30 per cent of total DRAM production in 2027, up from 20 per cent in 2026, with projections reaching 40 per cent by 2029.

Where Consumer Supply Went

AI data centres are projected to consume approximately 70 per cent of global DRAM supplies in 2026. That reallocation has left consumer GPU markets struggling with inventory shortages and price inflation that JP Morgan describes as "chipflation" - a rate of increase vastly exceeding economy-wide inflation measures. The Consumer Price Index for software and accessories rose nearly 25 per cent over two years, whilst video game consoles and laptops saw similar increases, contributing to an unusual contraction in PC shipments.

Regional markets outside the United States face even steeper premiums. In Japan, some gaming GPUs now sell at prices roughly 41 per cent higher than their American equivalents. Resale channels have amplified the problem; Sony's PlayStation 5 Pro, for example, trades at approximately $500 above manufacturer's suggested retail price through secondary sellers.

The shortage stems partly from overlapping technology transitions. Producers were already shifting from DDR4 to DDR5 memory formats when AI data centre construction accelerated, creating a dual constraint on manufacturing lines. Now, the industry is moving toward HBM4 and HBM4E specifications, further tightening supplies as production tooling is reconfigured.

Capacity Sold Years in Advance

Micron has reportedly sold 75 per cent of its 2027 HBM output, with CEO Sanjay Mehrotra stating in an earnings call that supply will be "much tighter in calendar 2027 and 2028" than in 2026. Those contracts were signed at prices "much higher than 2026 prices," according to Mehrotra. Samsung has committed nearly 80 per cent of its 2027 memory output, with roughly 70 per cent locked into five-year agreements, according to KB Securities, a Korean investment firm.

SK Hynix CEO Kwak Noh-jung described the coming year as potentially "the worst year in the industry's history from the supply perspective" in remarks to press in July. Chairman Chey Tae-won told attendees at an industry event that building adequate wafer supply will require "at least four to five years," with production lagging 20 per cent behind demand until at least 2030.

The concentration of production among three suppliers leaves little room for rapid capacity expansion. All three companies have announced facility investments, yet semiconductor fabrication plants require years to construct and bring online. Micron ceased direct-to-consumer sales in 2025 to focus resources on AI production, a decision that underscores the strategic priorities driving current allocation choices.

Divergent Views on Duration

Not all industry observers expect the shortage to persist through the decade. Acer CEO Jason Chen suggested to press that prices could plateau in 2027, citing potential competition from lower-cost Chinese memory producers. Questions also remain about whether AI infrastructure investment will sustain the growth rates that justify memory suppliers' current buildout plans.

A South Korean industry publication caused memory stocks to decline in October 2026 when it quoted analysts forecasting that a "flood of new supply" in 2029 could transform the memory market into a highly competitive environment. The analysis suggested that current capacity expansions might overshoot demand if AI adoption slows or if architectural efficiencies reduce per-workload memory requirements.

Nvidia CEO Jensen Huang dismissed the possibility of a supply surplus, noting that "one of the things that's really great about memory is that you can always use more of it." The comment reflects the data centre sector's view that memory remains a fundamental constraint on model scale and inference throughput, with demand likely to absorb new capacity as it comes online.

What Consumers Face

For consumer electronics buyers, the immediate outlook remains difficult. Prices for memory-intensive products - gaming GPUs, high-performance laptops, and next-generation consoles - are unlikely to moderate whilst data centre contracts absorb the majority of output. The industry's transition to HBM4 and HBM4E will occupy production lines through 2027, delaying any return of capacity to consumer-focused DRAM and DDR5 modules.

The pricing environment has already altered purchasing behaviour. PC shipments contracted in 2026, an unusual trend during a period without major economic recession, suggesting that consumers are delaying upgrades or shifting to lower-specification devices. That demand destruction may eventually influence manufacturers' allocation decisions, but multi-year contracts with hyperscale cloud providers and AI labs provide revenue certainty that consumer markets cannot match.

The longer-term question is whether memory producers' bets on sustained AI growth will prove correct. If architectural improvements or slowing adoption reduce HBM demand growth, capacity could return to consumer markets more quickly than current projections suggest. Until then, shoppers face a market shaped by priorities - and profit margins - that lie elsewhere.