TSMC's 2nm Process (N2) Explained: What It Means for Chips

TSMC's N2 is the company's first production-ready "2-nanometer-class" manufacturing process, and the thing that actually makes it a generational leap isn't the "2nm" marketing number — it's that N2 is the first TSMC node built entirely on gate-all-around (GAA) nanosheet transistors instead of the FinFET transistors every TSMC node back to 16nm has used. TSMC says N2 began volume production in the fourth quarter of 2025, with AMD's next-generation EPYC "Venice" server chip confirmed as the first high-performance computing product taped out on the node, and Apple, Nvidia and Qualcomm widely reported — though not yet officially confirmed by those companies — as early customers for phone and AI silicon.

What TSMC's N2 Process Actually Is

"N2" is TSMC's internal name for its 2-nanometer-class logic process. Like "7nm" or "5nm" before it, the number doesn't correspond to a literal, measurable 2-billionth-of-a-meter feature on the chip — it's a marketing-era holdover from when node names tracked transistor gate length. What N2 actually represents is TSMC's next full-node jump in transistor design, and the company has been explicit about what changed: N2 is TSMC's first high-volume node to use gate-all-around (GAA) nanosheet transistors, a structural shift the company first detailed publicly in 2022 when it unveiled N2 alongside its FinFlex standard-cell architecture at its North America Technology Symposium.

According to TSMC's original N2 announcement, the N2 platform was designed from the outset with multiple variants — a high-performance version, a mobile-compute baseline version, and chiplet integration options — rather than a single one-size-fits-all recipe, reflecting how differently a flagship AI accelerator and a phone SoC need to be tuned even on the same base process.

FinFET vs. Nanosheet: The Transistor Change That Defines N2

Every TSMC node from 16nm through N3 (3nm) used FinFET transistors, where the current-carrying channel is a vertical silicon "fin" with the gate wrapped around three of its four sides. It's an efficient design, but as fins get thinner to shrink chips further, controlling current leakage gets harder and the performance gains per node get smaller.

N2's nanosheet transistors solve that by replacing the fin with several stacked horizontal sheets of silicon, with the gate material wrapped completely around each one on all four sides — hence "gate-all-around." That full wraparound gives engineers much tighter control over the current flowing through the channel, which is what lets TSMC keep improving speed and power efficiency even as the physical dimensions barely shrink. It's the same fundamental transistor-architecture leap Samsung and Intel have also made on their own competing nodes, just arriving on TSMC's roadmap at N2 rather than one or two nodes earlier.

The nanosheet switch also changes how much design flexibility chipmakers get per cell. Because each nanosheet's width can be tuned independently, TSMC's NanoFlex standard-cell library lets designers mix taller, higher-performance cells with shorter, denser ones on the same die — putting speed where a chip's critical path needs it and density everywhere else. That flexibility is one reason TSMC positions N2 as suitable for everything from power-constrained phone chips to performance-hungry AI accelerators on the same base process, rather than needing entirely separate nodes for each.

N2 still delivers power through the front of the wafer, the same direction as every prior FinFET node. TSMC has said backside power delivery — moving the power-routing wiring to the underside of the chip, freeing up the front side for signal routing — arrives with later nodes such as A16 and A12's Super Power Rail, not with N2 itself. That staggering is deliberate: TSMC introduces one major structural change (the transistor) at N2, then layers the next major structural change (power delivery) on top a couple of nodes later, rather than shipping both risks in the same node at once.

When Did N2 Enter Production? TSMC's Own Timeline

TSMC first disclosed N2 in 2022 with a target of entering production in 2025. Three years later, the company's current logic-technology overview confirms that guidance held: "TSMC's 2nm (N2) technology has started volume production in 4Q25 as planned," built on what TSMC describes as its "industry-leading first-generation nanosheet transistor technology."

That status update matters because advanced nodes routinely slip. N2 didn't — at least according to TSMC's own account — which is one reason the company has been able to move straight into talking publicly about N2's derivatives (N2P and N2U) and the nodes that follow it (A16, A14, A13 and A12) rather than defending a delayed ramp.

Who Is Using TSMC's N2 First: Confirmed and Reported Customers

The one customer relationship TSMC and a chip designer have jointly, officially confirmed on N2 is AMD. In an April 2025 announcement, AMD said its next-generation EPYC server processor, codenamed "Venice" (built on AMD's Zen 6 architecture), had become "the first HPC product in the industry to be taped out and brought up on the TSMC advanced 2nm process." TSMC Chairman and CEO Dr. C.C. Wei confirmed the relationship directly: "We are proud to have AMD be a lead HPC customer for our advanced 2nm (N2) process technology and TSMC Arizona fab." The same announcement said 5th Gen EPYC chips (on an earlier node) had also been successfully brought up at TSMC's Arizona fab — a separate, U.S.-manufacturing milestone bundled into the same release.

Beyond AMD, the rest of the "who's on N2 first" picture is built on industry reporting rather than company statements, and should be read that way. Taiwanese and international trade press have reported that Apple is expected to use N2 for its A20-generation iPhone chips and M6-generation Mac chips, and that Nvidia is evaluating N2 or TSMC's follow-on A16 node for future Rubin Ultra and "Feynman" AI accelerators — reports that, as of this writing, conflict with each other on exactly which node Nvidia will use and when. AMD's own Instinct MI450 AI accelerator has also been reported, though not yet confirmed in an AMD press release, as an N2 part. None of Apple, Nvidia or Qualcomm has published its own newsroom statement naming N2 the way AMD did for Venice, so treat those specific chip-to-node pairings as credible but unconfirmed until the chipmakers say so themselves.

N2 vs. N3: Performance, Power and Density Compared

TSMC's clearest N2-vs-N3 numbers come from its original 2022 disclosure, and the company has repeated the same comparison basis since: versus N3, N2 delivers 10–15% higher speed at the same power, or 25–30% lower power at the same speed. Notably, TSMC's public materials describe N2 as a "full-node" improvement in performance and power efficiency over N3 but do not publish a specific logic-density percentage for N2 versus N3 — a contrast with how precisely the company quantifies density gains for later nodes like A14 and N2U (below). That's worth flagging rather than filling in with a guessed number.

N2 and Beyond: TSMC's 2nm-Era Roadmap

N2 isn't a single, static process — TSMC treats it as the base of a platform that keeps getting incremental variants for several years, the same pattern it used with N3/N3E/N3P. At its 2026 North America Technology Symposium, TSMC Chairman and CEO Dr. C.C. Wei said "TSMC's advanced process technologies lead the industry in density, performance and power efficiency," while the company detailed N2U — a 2nm-platform variant pitched as a "balanced option for AI, HPC, and mobile applications," offering 3–4% higher speed or 8–10% lower power than the existing N2P variant, with 1.02–1.03x the logic density, and slated for 2028 production.

Past N2 itself, TSMC's roadmap moves to the "A" naming series. A14, unveiled in April 2025, is positioned as a bigger step than N2U: up to 15% faster at the same power than N2, up to 30% lower power at the same speed, and more than 20% higher logic density, targeting 2028 production alongside a new NanoFlex Pro standard-cell architecture. A13, debuted at the 2026 symposium as "a direct shrink of its industry-leading A14 node," adds roughly 6% area savings with design rules TSMC says are fully backward-compatible with A14, for 2029 production. A12, previewed at the same event, pairs the A14 platform with Super Power Rail backside power delivery, also targeting 2029. TSMC has separately discussed an A16 node with backside power delivery of its own, though the exact production year for A16 has varied across TSMC's public disclosures and industry coverage, so it's best treated as "later this decade" rather than a fixed date pending TSMC's own confirmation closer to launch.

Why TSMC's 2nm Wafers Reportedly Cost More

TSMC has not published a specific N2 wafer price — pricing for TSMC's advanced nodes is negotiated privately with each customer and isn't part of its public technology disclosures. What's circulating is industry reporting, concentrated in Taiwanese financial press, that N2 wafers are priced in the neighborhood of $30,000 each, roughly 10–20% above N3's reported $25,000–$27,000 range and well below the 50% premium some early market chatter had speculated. Multiple outlets citing Taiwanese industry sources have also reported that TSMC is holding a firmer no-discount pricing line on N2 than on previous nodes, reflecting both tight capacity and surging AI-driven demand. None of this is TSMC-confirmed pricing — it should be read as "reportedly," sourced to Taiwanese trade press and analyst commentary, not as an official TSMC figure.

Whatever the exact number, a real price increase on leading-edge wafers flows downstream: it's one of the reasons flagship phone chips, AI accelerators and high-end PC silicon tend to get more expensive, not cheaper, with each new process generation, even as those chips also get faster and more efficient per watt — the same power-efficiency-versus-cost tradeoff that shows up when comparing thermal design power across chip generations.

N2 and the AI Chip Capacity Crunch

N2's arrival is landing in the middle of the broadest AI-driven demand surge TSMC has ever managed, and that's visible in how the company talks about capacity on its quarterly earnings calls: management has repeatedly described demand across advanced nodes and packaging as outstripping supply, with N2-family capacity expected to expand sharply through 2027 as TSMC brings more fabs — including sites in Hsinchu, Kaohsiung and its new Arizona campus — online for 2nm production. That capacity math is also tied to AI accelerators' growing appetite for high-bandwidth memory; chips built on N2 that pair with next-generation HBM4 memory are part of why both logic and memory supply chains are stretched simultaneously heading into 2027.

For buyers of finished devices, the practical effect shows up with a lag: N2 chips for AMD servers are expected in 2026, phone and PC silicon built on N2 — including whatever actually ships in Apple's and Qualcomm's next flagship lines, alongside chips like Qualcomm's current Snapdragon X2 Elite and Apple's M-series silicon on earlier nodes — tends to arrive in consumer products a year or more after a node enters volume production, as designs move through validation, packaging and device integration.

What's Next for TSMC's 2nm Node

The near-term story for N2 is less about the base process — which TSMC says is already in volume production — and more about which chips actually ship on it, and when. AMD's EPYC "Venice" is the one N2 product with a confirmed name and a confirmed foundry relationship; it's described as on track for a 2026 launch. Apple's and Nvidia's N2 plans remain, as of this writing, a matter of industry reporting rather than company confirmation, and will likely only firm up as those companies approach their own 2026–2027 product announcements. Meanwhile, TSMC's own roadmap keeps moving: N2U and A14 are both pointed at 2028, with A13 and A12 lined up for 2029, each promising the same formula N2 itself delivered — more speed or less power, without needing anyone outside TSMC to invent a new transistor shape in between.

Written byMara Lindqvist

Mara Lindqvist edits the hardware desk. She covers graphics cards, processors, memory and storage, the foundries and chip designers behind them, and what the numbers on a spec sheet mean for people choosing a PC. Specifications in her stories come from manufacturer spec pages and datasheets.

More from Mara LindqvistAll Hardware coverage