By Memorial Day 2026, the leading-edge foundry race has resolved into a structure the industry has not seen since the 28nm era: one dominant supplier with pricing power, one credible alternative for non-flagship work, and one IDM-turned-foundry burning cash to stay in the conversation.

TSMC began N2 volume production in Q4 2025 at Fab 22 (Kaohsiung), with Fab 20 (Hsinchu) ramping behind it. CEO C.C. Wei has called yields "good" — industry triangulation puts defect density already inside the band where N3 was 12 months into its ramp. Demand has overrun the initial 40,000 wpm plan. TSMC is now guiding to 100,000 wpm by year-end 2026 and 200,000 wpm by 2027 [Tom's Hardware, Nov 2025].

Samsung's SF2 hit 55–60% yield in late 2025 and its second-gen SF2P crossed the 70% mass-production threshold in January 2026 [TrendForce, Jan 2026] — the first time in three nodes Samsung Foundry has cleared that bar before a TSMC equivalent matured. Tesla's AI6 dojo-successor chip ($16.5B deal), DeepX, MicroBT, and Canaan are confirmed customers. Qualcomm and AMD are reportedly negotiating to dual-source portions of their 2nm roadmaps.

Intel 18A is producing Panther Lake compute tiles at Fab 52 (Arizona), but at yields the company itself describes as inadequate for target margins. Industry estimates put good-die yield at 20–25% in early 2026, improving roughly 7 percentage points per month — meaning industry-standard ~85% won't arrive until early 2027 [Tom's Hardware, Nov 2025]. The hyperscaler customers we cover in Article 1 (NVIDIA) account for the only confirmed external 18A wins of consequence.

The strategic question is no longer who has 2nm. It's who can afford to use it.

The era of "node leadership equals market leadership" is fracturing. TSMC's $30,000 N2 wafer — a 10–20% premium over N3's $25–27K, well below the 50% hike the rumor mill floated last summer [TechNode, Oct 2025] — is restrained pricing relative to TSMC's monopoly position. The implication: TSMC needs N2 to fill faster than N3 did. That tells you two things. First, customer concentration in HPC/AI is high enough that TSMC is self-limiting its take rate. Second, the next leg of pricing power moves to N2P and A16 in 2027, not N2. If your roadmap is locked to a single foundry for flagship silicon in 2027–2028, you are exposed to a step-function price reset. Samsung at 70% SF2P yield is, for the first time since the Snapdragon 8 Gen 1 fiasco, a defensible second source for non-bleeding-edge flagship work — particularly for customers who can tolerate a half-node performance gap in exchange for 15–20% wafer-cost relief and supply diversification.

Capex discipline at the leading edge has collapsed. TSMC's 2026 capex guide of $52–56B [DCD, Apr 2026] is a ~30% jump over 2025's $40.9B, with 70–80% earmarked for advanced nodes. Depreciation will compress gross margin by ~200 bps before mix tailwinds. If you're modeling fabless customers, build a wafer-cost step into 2027 numbers: N2 at $30K versus N3 at $25–27K is a ~15% bill-of-materials hit on a flagship SoC, and TSMC has signaled mid-single-digit price increases across N3/N5/N7 in 2026 [TrendForce, Oct 2025]. The capital return story at TSMC is intact, but the FCF inflection moves right by 12–18 months. For Intel, the math is harder: 18A at sub-50% yield through most of 2026 means the foundry segment burns cash regardless of Panther Lake volume. The hyperscaler 18A wins are symbolically critical but won't move the P&L in 2026.

Dual-sourcing at the leading edge is finally real, but the qualification timeline isn't. A Samsung SF2P qual cycle for a complex SoC remains 9–12 months minimum; if you don't have a parallel design today, you don't have a hedge for 2027. Three concrete actions:

N2 has confirmed what the equipment order book signaled 18 months ago: leading-edge foundry is now a one-and-a-half-supplier market. TSMC's pricing restraint is a tell — it needs volume more than margin on N2, because A16 and high-NA-enabled A14 are where the real economics sit. Samsung is a real second source for the first time in a decade, but only at SF2P and only for customers willing to absorb a qualification cycle. Intel's 18A is a technology win and a financial problem. The September Apple teardown is the moment the thesis is marked to market. Plan accordingly.

Silicon & Steel Intelligence Desk — Foundry & Advanced Logic. Brief errors or update intelligence: [email protected]