The earlier advanced-process gate gave an overview of 2nm, 3nm, and GAA; this piece focuses on TSMC’s 2nm (N2) itself: what it is, who’s using it, what the supply chain looks like, and where its rivals stand.
This is the deep-dive version of the foundry gate in The AI Hardware Supply Chain, End to End, zoomed in on TSMC’s 2nm. To make AI chips faster and more power-efficient, you ultimately have to build them on this kind of most-advanced process.
What N2 Is: TSMC’s First GAA Process
TSMC’s 2nm is called N2, and its biggest significance is this: it’s TSMC’s first volume-production node to switch to nanosheet/GAA (Gate-All-Around) transistors, taking over from the FinFET architecture used in the 3nm generation.
The difference is in how the transistor “turns off the tap.” FinFET’s gate wraps the channel on three sides; GAA fully encloses it on all four, giving better control over current and less leakage, and holding performance even at smaller dimensions. The performance figures are clearer for the enhanced N2P: TSMC officially says that, versus N3E, N2P delivers roughly 18% more speed at the same power, or 36% less power at the same speed, with logic density up to about 1.2x. To put it another way: FinFET is like a sluice gate walled off on three sides; GAA adds the fourth side too, so the water (leakage) is much less likely to seep out.
Timeline and Fabs: Mass Production in 2025, Home Turf in Taiwan
N2 entered mass production in the fourth quarter of 2025. On its first-quarter 2026 earnings call, TSMC said yields were good, with demand coming from both smartphones and HPC/AI. The production home turf is in Taiwan: Fab 20 in Hsinchu and Fab 22 in Kaohsiung, expanding capacity in multiple stages.
Looking ahead, the official timing for the enhanced N2P is volume production in the second half of 2026; the next generation A16 will bring in backside power delivery (Super Power Rail, which moves the power-delivery wiring to the back of the chip to free up space on the front). Official investor documents originally listed it for the second half of 2026, while later industry reporting suggests it may slip to 2027 — worth tracking in TSMC’s next official roadmap. As for the US Arizona fab, per public documents, 2nm and A16 won’t arrive until the end of this decade, so it isn’t the supply mainstay for 2026.
Who’s Using It: AMD and MediaTek Confirmed
TSMC has never disclosed its customer list, but some customers have confirmed it themselves. AMD’s next-generation Venice EPYC server processors already use TSMC 2nm and are in production ramp; MediaTek has also announced it completed the design tape-out of a flagship chip on TSMC 2nm/N2P, with volume production expected by the end of 2026.
The other names often mentioned — like Apple being first on N2, or NVIDIA and Qualcomm adopting 2nm or A16 — are currently mostly market reporting and analyst estimates, with no official confirmation yet, so when you read about them, keep “confirmed” and “market rumor” separate. What’s certain is that a most-advanced process like 2nm starts with limited capacity and high pricing, and the ones racing to be first are usually applications like phones and HPC/AI that are willing to pay a premium.
Pricing, Yield, and Capital Expenditure
When it comes to the numbers, you have to be very careful about which are official and which are market chatter.
TSMC officially only says N2 yields are “good,” without publishing a yield percentage, so the figures circulating outside shouldn’t be taken as fact. The figure of roughly US$30,000 per wafer is also market-reporting grade; TSMC has not disclosed pricing. What’s more certain is capital expenditure: TSMC’s official 2026 capex range is about US$52 to 56 billion, skewing toward the high end, mainly because of strong HPC/AI demand that keeps it pulling in equipment and expanding capacity. That alone shows just how much money it takes to scale an advanced process.
Where the Rivals Stand: Samsung and Intel
At the 2nm tier, TSMC isn’t fighting alone. Samsung Foundry began first-generation 2nm production in the fourth quarter of 2025 and is ramping its second generation in 2026. Intel’s 18A (using RibbonFET transistors plus PowerVia backside power delivery) is a 2nm-class rival that keeps improving yields and has begun shipping, with the key decision on the next-generation 14A landing in the second half of 2026 through 2027.
One technical difference that often comes up is backside power delivery: Intel 18A already uses PowerVia, while TSMC’s baseline N2 hasn’t brought it in yet — that waits for A16 with Super Power Rail. Whose yields hold steady, and who first gets a customer’s product to volume smoothly, will be the keys to this 2nm race.
The Supply Chain and the Role of Taiwanese Firms
The most critical equipment in N2 front-end manufacturing is still ASML’s EUV; that said, TSMC currently doesn’t treat the pricier High-NA EUV as a must for 2nm, and per reporting it doesn’t plan to bring it into volume production at least until the end of this decade.
As for the “2nm supply chain / theme stocks” often discussed in Taiwan’s market, analysts and the market will name a slate of firms: design services and silicon IP (such as GUC and Alchip), packaging and test (ASE and SPIL), substrates (Unimicron), photomask pods and clean consumables (such as Gudeng), advanced-process materials and reclaimed wafers, wet-process and equipment, and more. Here we need to be especially clear: these lists represent “possible supply-chain roles,” and many come from brokerage and media speculation; being named doesn’t mean a firm has won N2 orders, nor does it indicate the degree of benefit. This article only describes industry roles; it does not compile beneficiary stocks, does not rank individual names, and does not constitute investment advice.
Key Takeaways for This Gate
N2 is TSMC’s pivotal step from FinFET to GAA: mass production at the end of 2025, ramping volume in 2026, with home turf in Hsinchu and Kaohsiung in Taiwan, while the 2nm at the US Arizona fab won’t arrive until the end of this decade.
On the customer side AMD and MediaTek are confirmed, while Apple and others are market rumor; the pricing and yield figures are mostly market estimates, and TSMC only says yields are good. Rivals Samsung and Intel are both chasing, with backside power delivery being a frequently compared technical point. The list of Taiwanese supply-chain firms is mostly analyst speculation — just understand the roles.
To see the overview of 2nm/3nm/GAA, go back and read What Is the Advanced Process; to see the full picture of foundry, read foundry; to see the lithography equipment behind it, read EUV; to look back at the whole chain, head back to the supply-chain overview.