Why Memory Shortages Could Persist Until 2030

DRAM Supply Outlook

According to a Nikkei Asia report, even if major DRAM manufacturers increase output, global supply is projected to meet only about 60% of demand by the end of 2027. The chairman of SK Group warned that the shortage could persist until 2030.

Most of the new fab capacity will not become operational until 2027‑2028. The only new line expected in 2026 is SK Hynix’s wafer plant in Cheongju. To satisfy market demand, production would need to grow roughly 12% per year in 2026‑2027, but Counterpoint Research shows current plans target a 7.5% annual increase, creating a sizable supply‑demand gap.

AI‑Driven Allocation to HBM

AI workloads are causing the newly built capacity to be allocated primarily to high‑bandwidth memory (HBM), which offers higher profit margins and a clearer demand profile than ordinary DDR5 DRAM used in PCs and smartphones. A previous report highlighted that large AI models have a “bottomless” memory appetite and that HBM sales to AI data centers command multiples of the price of consumer DRAM. OpenAI and other AI leaders have reportedly secured nearly half of global DRAM monthly production.

Consequently, even when the new factories come online, the majority of output will be dedicated to HBM for AI, further squeezing the limited DRAM capacity available for consumer electronics. This structural shortage means AI data centers obtain more, newer, and more expensive memory, while the supply for phones, laptops, VR headsets, and gaming handhelds becomes increasingly constrained.

Software Compression Limits

Google’s TurboQuant compression algorithm reduces the memory footprint of large models by more than six times without precision loss, using a two‑stage compression process. The announcement triggered a brief rally in memory‑related stocks, but the algorithm cannot close the physical capacity gap created by the massive AI expansion.

Consumer Impact

The shortage manifests as higher device prices, potential “silent downgrades” where manufacturers reduce default memory sizes, disrupted upgrade cycles (e.g., the shift from 16 GB to 32 GB may stall), and possible delays or cancellations of less profitable product lines.

Reference links:

https://asia.nikkei.com/business/tech/semiconductors/memory-shortage-set-to-run-until-2027-as-chipmakers-focus-on-ai

https://www.theverge.com/ai-artificial-intelligence/914672/the-ram-shortage-could-last-years