Rapidus Targets 2027 for 2nm Chip Mass Production Amid Customer Uncertainty
Rapidus’s 2nm Fab: A Nation’s Bet on Semiconductor Sovereignty
The Tech TL;DR:
- Rapidus aims for 2027 mass production of 2nm chips using IBM IP, but no volume customers have committed yet.
- The single Hokkaido fab carries significant execution risk, with Japan’s government holding a golden share.
Rapidus has set a 2027 target for 2nm mass production at its single Chitose fab, but no high-volume customers have signed agreements. The project, backed by ¥267.6 billion in funding, hinges on a 2nm gate-all-around design derived from IBM’s 2021 node, with production flow optimized for AI-driven yield learning. However, the lack of confirmed customers and the technical challenges of High-NA EUV for 1.4nm loom large.
Why the 2nm Node Matters for Semiconductor Sovereignty
Rapidus's 2nm process, based on IBM's 2021 architecture, uses gate-all-around nanosheets.
Comparisons to TSMC’s N2 node, which moved to volume production in late 2025, highlight Rapidus’s gap. Rapidus’s current 2nm chips are projected to cost 10x more than Japan’s mainstream parts, per TrendForce.
The Single-Fab Bet: Risks and Renewable Advantages
Rapidus's reliance on a single fab in Chitose, Hokkaido, creates operational vulnerability. The fab's siting in Chitose offers abundant water, a cool climate, and renewable-energy potential across wind, solar, and hydro.
Japan’s government holds a golden share in Rapidus, allowing veto power over tech partnerships. This structure, enabled by a 2025 regulatory change, converts to a controlling stake of roughly 60% if performance targets fail. The state’s planned government backing is about ¥2.9 trillion, with ¥150 billion in equity added in early June.
Customer Conundrums: 60 Talks, No Volume Orders
Atsuyoshi Koike claims over 60 companies are in discussions, but only 10 have received quotes. TrendForce reports IBM and Tenstorrent as potential clients, while Fujitsu evaluates 1.4nm CPU outsourcing for its Fugaku supercomputer. However, no high-volume agreements exist, leaving Rapidus exposed to fixed costs if the 25,000-wafer/month ramp falters.
“Rapidus’s cost model depends on filling the 25,000-wafer ramp, and a fab running well below capacity carries the same fixed depreciation as a full one.”
TSMC’s JASM Venture: A Contrasting Strategy
Japan’s dual-track approach includes TSMC’s JASM venture in Kumamoto, which began mass production in December 2024 on mature 12nm, 16nm, 22nm, and 28nm nodes. JASM’s second Kumamoto fab is targeting production around 2028 on 3nm. Unlike Rapidus, JASM leverages TSMC’s proven processes, while Rapidus bets on domestic innovation.
The contrast highlights Japan’s risk tradeoff: JASM offers immediate output, while Rapidus’s 2nm/1.4nm roadmap depends on High-NA EUV, which imec’s position is that 1.4nm single-patterned layers require.