iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
TSMC plans to build another 3nm semiconductor fabrication plant in Tainan, Taiwan, with production targeted for the first half of 2028, according to The Information’s October 15, 2026 briefing. The project responds to strong customer demand as artificial-intelligence systems require increasing amounts of leading-edge computing silicon. It is part of a broader expansion that also includes 3nm manufacturing in Arizona and Japan.
What TSMC has announced
The newly reported facility will be located in Tainan, alongside TSMC’s existing advanced-node operations in Taiwan. The Information attributes the first-half-2028 production target to strong customer demand. The report is paywalled, so the location, timing and rationale should be treated as attributed reporting rather than as details independently published in a new TSMC announcement.
TSMC has not publicly stated the new fab’s exact wafer capacity, named customers or product allocation in the material available for this report. Those omissions matter: a fab’s eventual output depends on equipment installation, process qualification, customer tape-outs and the mix of products approved for high-volume manufacturing.
Recommended Free Tools
Why artificial intelligence is driving another 3nm build
More computation per AI model
TSMC said in its April 17, 2025 earnings call that it continued to see robust AI-related demand, reaffirmed its expectation that AI-accelerator revenue would double in 2025, and planned to double its CoWoS advanced-packaging capacity during that year. The company uses “AI accelerators” to mean AI graphics processors, application-specific integrated circuits and high-bandwidth-memory controllers used for data-center training and inference.
In its 2026 outlook, TSMC said the adoption of AI models by consumer, enterprise and sovereign customers is increasing the need for computation and supporting demand for leading-edge silicon. The company also pointed to reasoning models as a potential source of wider AI adoption. Larger or more capable models generally require more accelerator compute, memory bandwidth and advanced packaging, which increases demand for wafers made on advanced process nodes.
AI demand is broader than one chip category
“AI orders” do not refer only to a single type of processor. Data-center systems can combine general-purpose CPUs, AI GPUs, custom AI ASICs, networking silicon and HBM controllers. TSMC’s definition covers several of these accelerator components, while its comments about leading-edge demand address the broader silicon needed to build and deploy AI systems.
Rank #2
TSMC’s 3nm expansion by location
The Tainan project is one element of a multi-region capacity plan. The schedules below distinguish facilities already producing chips from projects that are planned or being accelerated.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
| Facility | Location | Process and application context | Public timing or status |
|---|---|---|---|
| New Tainan fab | Tainan, Taiwan | 3nm; customer allocation and capacity not stated | Production targeted for the first half of 2028, according to The Information on October 15, 2026 |
| Arizona Fab 2 | Arizona, United States | 3nm high-volume manufacturing | TSMC’s 2025 annual report says the schedule is being pulled forward to the second half of 2027 |
| JASM second fab | Kumamoto, Japan | 3nm planned for AI-driven demand | Planned; TSMC’s annual report does not provide a production date in the cited material |
| Fab 18 GIGAFAB | Southern Taiwan Science Park, Taiwan | 3nm volume production | Entered volume production before TSMC’s December 29, 2022 announcement |
TSMC also said it completed a second large land purchase in Arizona to preserve flexibility for strong, multi-year AI-related demand. That purchase is not itself a promise that every parcel will become a 3nm fab; it gives the company room to adjust the scale and timing of later construction.
How large is the overall capacity push?
TSMC reported NT$3,809.05 billion in 2025 revenue, up 31.6% from the prior year. Technologies at 7nm and more advanced accounted for 74% of wafer revenue. Across facilities managed by TSMC and its subsidiaries, the company reported more than 17 million 12-inch-equivalent wafers of annual capacity.
At its first-quarter 2025 call, TSMC described a planned total U.S. investment of US$165 billion after adding US$100 billion to the Arizona expansion. That figure covers the broader U.S. program, not just the Arizona 3nm line, and should not be interpreted as the cost of the new Tainan fab.
Rank #4
What 3nm adds compared with 5nm
TSMC’s December 29, 2022 technology comparison says its 3nm process can provide up to 1.6 times the logic density of 5nm and reduce power consumption by 30% to 35% at the same speed. “Up to” is important: actual results vary by chip design, libraries, memory configuration, clock targets and manufacturing options.
Higher density can fit more logic into a given die area or enable a smaller die for a similar design. Lower power at an equivalent performance target can help data-center operators manage energy and cooling, while mobile-device designers can trade some of the gain for longer battery life or additional performance. These are process-level capabilities, not a guarantee that every 3nm product will achieve the maximum figures.
Best Value
Which companies are ordering TSMC’s 3nm chips?
The cited announcements do not identify the customers behind the new Tainan fab or provide a company-by-company order list. TSMC’s public descriptions instead identify demand categories: AI GPUs, custom AI ASICs, HBM controllers and other leading-edge silicon used in training and inference systems, as well as established smartphone and high-performance-computing applications.
Without a named customer disclosure, it would be misleading to assign the plant to a particular chip designer or to claim that a specific company has reserved a stated number of wafers. Customer commitments can also change as products move from design to qualification and as TSMC balances capacity among regions and nodes.
What the 2028 target means for chip buyers
It adds future supply, not immediate relief
A first-half-2028 production target is several manufacturing stages away from usable volume for most products. TSMC must complete construction, install and qualify equipment, bring the process into stable production, and then qualify individual customer designs. The new fab therefore addresses expected multi-year demand rather than instantly easing current allocation constraints.
Capacity will be distributed across a portfolio
TSMC is simultaneously expanding in Taiwan, accelerating Arizona Fab 2 and planning 3nm at the second Kumamoto fab. Geographic diversification can give customers more options for supply-chain resilience, but the three sites will not necessarily offer identical yields, packaging connections, product qualifications or ramp rates. A customer choosing a site will still need to match its design and logistics requirements to the qualified process available there.
Quick Recap
What to watch next
- Whether TSMC issues a direct announcement confirming the Tainan fab’s construction scope, capacity and schedule.
- Progress toward Arizona Fab 2’s stated high-volume manufacturing target in the second half of 2027.
- A production timetable for the second Kumamoto fab and confirmation of which 3nm variants it will support.
- Updates on AI-accelerator demand, CoWoS capacity and the pace at which reasoning-model deployments translate into wafer orders.
- Customer qualification announcements, which would provide more reliable evidence than broad references to AI demand about who is using each site.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
