On July 3, Micron broke ground on a $9.3 billion (1.5 trillion JPY) fab expansion in Higashi-Hiroshima, Japan – one of the largest single-site memory capex investments globally. The goal: ramp HBM advanced capacity for AI servers and close the gap with SK Hynix.

The Hiroshima site, acquired via Elpida in 2013, is Micron’s core DRAM and HBM production base. The new fab will mass-produce next-gen 1δ-process DRAM and HBM4E, with Micron already shipping HBM4 for NVIDIA’s Vera Rubin platform in volume. Executives confirmed the line will cover post-1γ advanced nodes and next-gen HBM stacks. Equipment installation begins H2 2028, with commercial shipments starting that summer and gradual ramp to full capacity.
According to Counterpoint, SK Hynix led HBM with 58% share in Q1 2026, followed by Samsung and Micron at 21% each. Micron has climbed from 9% at end-2024 to overtake Samsung. With this expansion, Micron aims to reshape the HBM supply duopoly and capture more AI memory orders long-term.
The Japanese government is backing the project with up to ¥500 billion in direct subsidies plus R&D support – total ~¥775 billion, covering nearly half the cost. This is part of Japan's broader semiconductor revival, alongside TSMC Kumamoto and Rapidus. Hiroshima prefecture expects over 1,000 new local jobs.
Industry-wide, AI memory is in a super-cycle. Bank of America forecasts $54.6 billion global HBM market in 2026, up 58% YoY. Micron CEO Sanjay Mehrotra stated at the ceremony: "Global memory demand is entering unprecedented growth – this capacity positions us for long-term AI hardware orders."
From ICgoodFind: Japan's subsidies + Micron's HBM push = a real three-way race. Watch for who yields first on HBM4E – that's the next market share pivot.
