Taiwan Semiconductor Manufacturing Company is accelerating capacity expansion for its 2nm process and aims to reach a monthly output of 120,000 wafers by the end of the year, according to a report carried by Futu Niuniu citing surging AI demand as the driver.

The 120,000-wafer figure marks an aggressive ramp for TSMC's most advanced node. It signals that demand projections from AI accelerator designers and smartphone customers have firmed enough for the foundry to commit additional tooling and cleanroom capacity ahead of its original schedule.

AI is the central force behind the acceleration. Training and inference clusters have absorbed advanced-node supply faster than most forecasts anticipated, and foundry capacity at the leading edge remains the tightest link in the AI supply chain. For TSMC, that demand translates directly into capacity commitments: customers appear willing to reserve 2nm volume well in advance, giving the company the volume visibility needed to justify a faster buildout.

The 2nm node, which TSMC designates N2, moves the company from FinFET transistors to gate-all-around architecture. That transition demands new process steps, new toolsets, and requalified yields — all of which make a 120,000-wafer-per-month target a substantial engineering undertaking, not merely a procurement exercise. Reaching it by year-end implies that yield learning at the node is progressing in line with the company's internal schedule.

Hitting the target also carries competitive weight. Samsung Foundry and Intel are both racing to qualify their own sub-3nm-class technologies for high-volume customers, and every quarter of a clean ramp widens TSMC's lead at the leading edge. Conversely, any slippage in the 2nm ramp would tighten advanced-node supply further and push major AI chip designers toward alternative sources or older, more available nodes.

For TSMC's customers, the accelerated timeline offers relief. AI accelerator suppliers have faced allocation constraints at the most advanced nodes, and additional 2nm capacity coming online by year-end would ease those queues — though pricing at the leading edge is expected to remain premium, as it has through every recent node transition.

The report does not break down how TSMC will distribute the 120,000 monthly wafers across its fabs or how the figure splits between phases of the ramp. What it does establish is direction: AI demand is strong enough that the world's largest foundry is pulling its most advanced capacity plans forward, and investors should watch quarterly capacity disclosures to confirm whether the year-end target holds.