Micron warns global memory shortage will continue through 2026

Micron Technology has reported that the global memory chip shortage has intensified over the past quarter, driven by increased demand for high-end semiconductors used in AI infrastructure.

The company’s executive vice president, Manish Bhatia, described the shortage as “unprecedented” and expects it to persist beyond 2026.

High-bandwidth memory needed for AI accelerators is limiting supply for traditional markets like phones and PCs, with automakers and robotics also increasing demand.

Micron is expanding manufacturing in the US and Asia, including a US$100 billion plant near Syracuse, New York, and plans to produce 40% of its DRAM in the US.

The company also announced a US$1.8 billion investment in a Taiwan site to accelerate production, with wafers expected to begin in late 2027.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

The ‘unprecedented’ memory shortage has specific numbers that need to be checked

An “unprecedented” shortage has been described by Micron and its executive vice president of operations, with higher memory prices being cited. The exact size of those increases should be verified.
Fourth-quarter DRAM contract prices were forecast to rise by over 75% year over year, with total device costs projected to increase 8–10% in 2025 and 5–7% in 2026 1.
The profit incentives behind memory suppliers’ output choices should be reviewed.
The price gap between High Bandwidth Memory (HBM)3e and server DDR5 (a common server memory standard) is expected to shrink from 4–5x to 1–2x by the end of 2026 2.
If DDR5 margins are viewed as higher than HBM3e, manufacturing lines could be shifted from HBM to conventional DRAM. Future supply for phones, PCs, and data centers could be affected 3.

The memory crunch creates a clear opportunity in advanced packaging

The HBM squeeze opens room for suppliers tied to advanced packaging 4.
HBM needs advanced packaging such as Chip-on-Wafer-on-Substrate (CoWoS) to connect with AI accelerators. Tight CoWoS output drives much of the bottleneck 4.
Reports say TSMC’s CoWoS lines are fully booked, with Nvidia reserving more than half 5.
That spillover lifts Outsourced Semiconductor Assembly and Test (OSAT) firms, companies that package and test chips for chip designers and manufacturers 6.
ASE is projected to lift CoWoS throughput to 20,000–25,000 wafers per month by year end. Amkor is building a new $7 billion facility in Arizona for customers including Apple and Nvidia 6 7.