TSMC Starts 2nm Volume Production, Already Sold Out

TSMC has begun volume production of its 2nm-class N2 process, the first node it builds on gate-all-around nanosheet transistors, and the entire 2026 output is already sold out. Apple has reserved more than half of the initial capacity, with Qualcomm, MediaTek, AMD and Nvidia taking the rest, and most of the demand this time is coming from high-performance computing rather than phones. When the world's leading foundry ships a new node fully booked before the first wafers reach customers, it is the clearest signal there is that the AI compute boom now sets the pace of the entire semiconductor industry.

• N2 is TSMC's first gate-all-around (GAA) nanosheet node, delivering roughly 10-15% more performance at the same power or 25-30% lower power, plus about 15% higher density versus N3E.
• Capacity is fully booked for 2026, ramping across five fabs (two in Hsinchu, three in Kaohsiung) toward a target near 100,000 wafers per month.
• Apple takes more than half of the first wave; other customers include Qualcomm, MediaTek, AMD and Nvidia, with the majority of demand in HPC.
• TSMC expects 2nm capacity to grow at roughly a 70% annual rate through 2028, with N2P and the A16 variant following in the second half of 2026.

What is different about the N2 node?

N2 marks the moment TSMC retires the FinFET transistor design it has used for a decade and moves to gate-all-around nanosheet transistors. The distinction is physical: in a FinFET the gate wraps a channel on three sides, while a GAA nanosheet surrounds the channel completely, giving the transistor far better control over current leakage as features shrink. That control is what lets N2 deliver roughly 10 to 15% more performance at the same power, or 25 to 30% less power at the same performance, along with about 15% higher density than the previous N3E node. Those are generational gains, and they arrive exactly when AI accelerators are desperate for more compute per watt.

RelatedNvidia's RTX Spark Superchip Puts an AI PC in a Box

Why is the capacity already sold out?

A new leading node is always constrained, but this one is booked solid for all of 2026 before it has shipped in volume, and the reason is a shift in who is buying. Historically TSMC's newest process went first to mobile chips from Apple and Qualcomm. This time the majority of N2 demand comes from high-performance computing, the CPUs and AI accelerators that fill data centers. Apple still takes more than half of the initial capacity, but AMD and Nvidia are in the queue alongside it, and their appetite is effectively unlimited because every extra wafer becomes another AI chip a hyperscaler will buy. When demand outruns a fab's physical output, allocation, not price, becomes the bottleneck.

How fast can TSMC scale it?

The ramp is aggressive by TSMC's own standards. Five fabs are entering volume production this year, two in Hsinchu and three in Kaohsiung, and the company is targeting output near 100,000 wafers per month, with first-year N2 volume running about 45% higher than 3nm managed in its debut. Beyond that, TSMC expects 2nm capacity to grow at roughly a 70% compound annual rate through 2028, and it is already lining up the N2P refinement and the A16 variant for the second half of 2026. Even so, building fabs takes years and billions, so the supply will trail demand for the foreseeable future.

N2 versus the node it replaces

Who wins and who waits?

Being first in line is now a genuine competitive weapon. Apple's grip on more than half of early N2 output means its next chips get the density and efficiency edge before anyone else, and the AI chipmakers that secured allocation can promise customers better performance per watt than rivals stuck on older nodes. Everyone else waits, and waiting has a cost measured in product roadmaps. The concentration also underlines how much of the modern economy funnels through a single company in Taiwan: when one foundry's newest process is sold out, the pecking order for the world's most advanced chips is effectively set by who booked capacity first.

RelatedQualcomm's AI200 and AI250 Chase Nvidia in Inference

Our take

A sold-out leading node is the semiconductor world's version of a stadium selling out before the doors open, and it tells you more about the AI boom than any single chip launch. The move to gate-all-around transistors is a real generational step, and TSMC timed it perfectly for a market that will absorb every wafer it can make. The strategic story underneath is concentration: the most advanced computing on earth depends on one foundry's ability to ramp five fabs on schedule, and on Apple, Nvidia and AMD racing to lock in allocation. N2 is a triumph of execution. It is also a reminder of how narrow the base of the AI pyramid really is.

Original analysis by GenZTech. Reporting informed by Tom's Hardware.