2nm Capacity Sold Out Before Mass Production Even Began: Where TSMC, Intel, Samsung and Rapidus Stand

The 2nm generation is the current frontier of semiconductor miniaturization — the process node at the leading edge of manufacturing technology. The smaller the number, the more transistors fit in the same area, and the better the performance per watt. Four camps are now racing for that frontier from completely different starting positions.

Why is 2nm being fought over?

2nm is being fought over because AI chips and flagship smartphone chips are competing for the same leading-edge capacity. The main buyers of leading-edge processes used to be smartphone SoCs (chips that integrate a device's main functions). Then the enormous demand from AI data centers arrived on top, turning limited capacity into a contest among multiple giant customers.

The result is "sold out before it is built." TSMC's N2 (2nm generation) only entered volume production in December 2025, yet its 2026 capacity is reported as effectively sold out to major customers including Apple, NVIDIA, Qualcomm and AMD (SemiWiki).

TSMC keeps piling on investment to meet the demand. Capital expenditure rose from US$29.8 billion in 2024 to US$40.9 billion in 2025, and is planned at a record US$52–56 billion for 2026 (SemiWiki, TSMC) — a range consistent with the budget approved by the company's board. Why leading-edge fabs have become this expensive is explained in the economics of chip fabs that now cost over US$20 billion per building (in Japanese).

Where do the four camps stand right now?

The whole landscape fits in one table.

TSMC — the only camp that is both in production and sold out

TSMC is the only one of the four with both volume production and commercial success in hand. It began high-volume N2 production at two Taiwan fabs in December 2025, and the 2026 allocation is reported as effectively sold out (Tom's Hardware, SemiWiki). The roadmap runs unbroken: the enhanced N2P, then A16 — which adds backside power delivery, routing power from the back of the wafer so the front side can be dedicated to signals — and then A14 in 2028 (Tom's Hardware). The technology inside N2 is covered in depth in TSMC's 2nm generation — the structure that cut power by up to 35% at the same speed (in Japanese).

Intel — the technology is ready; the test is the volume ramp

Intel's 18A began production in November 2025 — but on the development line in Oregon, not the volume line in Arizona (Tom's Hardware). The technology is running; the move to large-scale production is the real test still ahead. 18A pairs a GAA structure with an early adoption of backside power delivery, which routes power from the back of the wafer so the front can be dedicated to signals. The next 14A generation targets production readiness in 2027–28 and plans the industry's first partial use of High-NA EUV, the newest class of lithography tools (Tom's Hardware). The full 18A story is in Intel's 18A process — two innovations and the watershed for its manufacturing comeback (in Japanese).

Samsung — fastest into production, walled in by yield

Samsung actually led on the mobile production calendar: its SF2 (2nm, GAA) has been producing mobile SoCs since mid-2025 (Tom's Hardware). But as of April 2026, its 2nm yield — the share of manufactured chips that work — was reported at roughly 55%, below volume-production levels (TrendForce). With yield that low, the more you build, the worse the economics get. Samsung plans HPC and automotive derivatives for 2026–27 and a finer 1.4nm-class node for 2027 (Tom's Hardware). The profitability gap with TSMC is analyzed in Samsung's foundry profitability slipping to 2028 and the 2nm yield gap (in Japanese).

Rapidus — Japan's late entrant targeting fiscal 2027

Rapidus is the last of the four to start. But it has already demonstrated Japan's first working 2nm GAA transistors on 300mm wafers — clearing the first gate of "can it be built at all" (Rapidus). Its mass-production target is fiscal 2027, backed by cumulative government support of roughly ¥2.4 trillion. The background to that public funding and the challenges that remain are covered in Japan Has Committed ¥2.35 Trillion to Rapidus — and Money Was Never the Hard Part.

Why did every camp switch to GAA?

At the 2nm generation, all four camps — TSMC, Intel, Samsung and Rapidus — switched their transistor structure to GAA (gate-all-around), in which the gate wraps the current channel on all four sides. The previous FinFET structure, which raises the channel as a fin and controls it from three sides, was losing its grip on current control as features kept shrinking.

Because GAA surrounds the channel from every direction, it controls current better than FinFET. The 2nm generation is therefore not just another number: it marks the first change of the transistor's basic structure in a decade. Intel's RibbonFET is also a form of GAA — different names, but all four camps converged on the same structural principle.

With the structure now common to everyone, the axis of competition has shifted from "which structure do you choose" to "who can mass-produce the same structure at the highest yield." As Samsung's reported ~55% shows, what separates winners from losers is not the blueprint but manufacturing execution (TrendForce).

How long will the shortage last?

The 2nm supply crunch is reported likely to last until around 2027 (SemiWiki). Demand from AI chips and flagship smartphone chips shows no sign of fading, while for now the only company that can actually add supply is TSMC, the one that has succeeded at volume production.

Two things would loosen the market. First, TSMC's record capex — US$52–56 billion in 2026 — coming online as new capacity. Second, the chasers becoming real alternatives: Intel transitioning to volume production in Arizona, Samsung lifting its yield, and Rapidus reaching its fiscal 2027 production start. Only then do buyers get genuine choice.

Until then, pricing power stays with the lead foundry. The four-way race over 2nm is not a technology beauty contest — it is the contest that decides who allocates the computing resources of the AI era. Keep this map at hand, alongside the deep-dive articles on each camp, as the race unfolds.