TSMC (Taiwan Semiconductor Manufacturing Company), the undisputed "absolute champion" of the global semiconductor industry, is moving to further expand its next-generation, cutting-edge "2-nanometer (2nm)" manufacturing footprint.
According to a report on November 25, 2025 by Taiwan's leading media outlet Liberty Times, TSMC is reportedly moving forward with plans to build three additional 2nm fabs domestically in Taiwan, on top of its existing construction plans.
A "NT$900 Billion" Additional Investment in 2nm Process: The Full Picture
First, let's organize the core facts of this report. TSMC is currently looking to substantially revise and expand its existing capital expenditure plans amid a surge in semiconductor demand driven by the artificial intelligence (AI) boom.
Plan Overview and Site Strategy
According to the report, TSMC is considering an additional investment on the following scale.
- Additional fabs to be built: 2nm process-capable fabs × 3 units
- Estimated investment: Approximately NT$300 billion per fab, for a total of approximately NT$900 billion (roughly ¥4.4 trillion / $30 billion)
- Leading candidate site: A specific development zone near the "Southern Taiwan Science Park (STSP)" in Tainan City (Block A, approximately 40 hectares)
This site selection is highly strategic. TSMC already has clusters of cutting-edge fabs (including Fab 18) concentrated in STSP, allowing for shared supply chains and infrastructure. The aim appears to be maximizing ramp-up speed by leveraging the existing ecosystem.
The Shocking Reality: Existing Plans "Are Not Nearly Enough"
What's crucial here is that this is not a "replacement" of the original plan, but purely a "net addition."
Until now, TSMC had planned to mass-produce 2nm chips under the following framework:
- Hsinchu Science Park (Baoshan): 2 fabs
- Kaohsiung: 5 fabs
- Total: 7 fabs
Normally, seven cutting-edge fabs should be more than enough to cover global demand. Yet this report suggests that even this is "not enough." This move toward a total of "10 fabs" is a scale that overturns conventional wisdom in the semiconductor industry, illustrating just how insatiable the appetite for computing resources has become in the AI era.
Why Is Further Additional Investment Needed Now?
Why is TSMC rushing to expand at such a rapid pace, even shouldering enormous risk in the process? Behind this lies a structural factor that can't simply be dismissed as "rising demand."
1. The "Miscalculation" of AI Demand and C.C. Wei's Prophecy
TSMC Chairman and CEO C.C. Wei has long emphasized the strength of AI demand, but it appears that events are unfolding at a pace that even exceeds the company's own expectations.
According to the report, Wei has expressed the following views in recent speeches and interviews:
- "AI-driven demand has reached roughly three times TSMC's existing capacity."
- "Capacity is 'not enough, not enough, still not enough.'"
The report also mentions an anecdote of NVIDIA CEO Jensen Huang personally visiting to plead "give us more chips," but this is merely the tip of the iceberg. Beyond AI accelerators (GPUs/NPUs) for data centers, the full-scale rollout of AI implementation in edge devices (PCs, smartphones) is intensifying the scramble for cutting-edge process capacity.
2. Apple, the "Massive Gravitational Force"
Another decisive factor is the behavior of Apple, TSMC's largest customer.
Apple is said to have already secured "more than half" of the initial allocation of TSMC's 2nm production capacity for the next-generation chips (A20 and A20 Pro) slated for the 2026 "iPhone 18" series.
This creates the following structural problem:
- Apple: Secures top priority allocation (over 50% of the total).
- Remaining allocation: NVIDIA, AMD, Qualcomm, MediaTek, and others fight over what's left.
For major mobile SoC players like Qualcomm and MediaTek, securing 2nm allocation is a matter of survival. With the existing plan (7 fabs) alone, capacity could be entirely consumed by Apple and NVIDIA, potentially forcing out other customers. This "3 additional fabs" move can be analyzed as a response to the intense pressure from these customers.
TSMC's Moat: Unmatched by Rivals
In analyzing this case, one cannot avoid the question: "Why don't customers simply go to Samsung or Intel instead?"
Recent reports have touched on Samsung's plans to produce AI chips for Tesla using its 2nm (SF2) process, as well as Intel's aggressive push with its "18A" process. However, the reality of the industry is harsh.
The High Wall of "Risk Cost"
According to sources within the supply chain, the high "risk cost" of switching to another foundry is a major deterrent.
In cutting-edge processes, the transition from design to manufacturing requires enormous cost and time. If yield rates fail to improve, or performance falls short of specifications, it directly translates into product launch delays and diminished competitiveness.
- Samsung: May not have fully shaken off the lingering image of yield and heat-generation issues from past processes (particularly its early GAA transition).
- Intel Foundry: Is still in the process of building a track record as an external customer-facing service (IFS), and has yet to demonstrate the same level of "certainty" as TSMC.
As a result, customers are left with little choice but to pick "TSMC, which can reliably deliver working chips even at a higher cost and with longer waits." This "monopoly on trust" is precisely the source that is enabling TSMC to commit to a NT$900 billion investment.
Geopolitics and Industrial Policy: A Return to Taiwan
This news also carries interesting implications in a geopolitical context.
Balancing with U.S. Investment
TSMC is also making a massive investment (up to $65 billion) in Arizona, USA, where it is building three fabs. At one point, concerns arose within Taiwan that technology and capital might flow out to the U.S., potentially hollowing out Taiwan's domestic industry (the so-called "de-Taiwanization" argument).
However, this latest report sends a clear message on the following points:
- The cutting edge remains in Taiwan: Even as the U.S. fabs come online, R&D and the largest-scale mass production hubs will remain in Taiwan.
- Not "Taiwan Plus One," but "Taiwan Expansion": Overseas expansion is merely complementary; the pace of investment in Taiwan itself is not slowing—if anything, it's accelerating.
C.C. Wei's remark—"All AI customers turn to us. There's no impact on Taiwan; if anything, we need the government to help us find more land"—underscores that Taiwan's status as the mother fab remains unshakeable.
The Future Technology Roadmap: Beyond 2nm
The report also touches on the "A14 (equivalent to 1.4nm)" process, which lies even beyond 2nm.
- 2nm (N2): Full adoption of nanosheet transistor (GAA) technology. Significant improvements in power efficiency and performance are expected. According to reports, some initial mass production activity (or risk production) has reportedly already begun this quarter (Q4 2025), with preparations underway for full-scale rollout in 2026.
- A14 (1.4nm): A next-generation site under construction in Taichung. This will introduce a process that further miniaturizes beyond 2nm.
By building a large number of 2nm fabs, TSMC is aiming to completely dominate AI semiconductor supremacy over the next 5 to 10 years—not just in terms of technology, but also in terms of physical production capacity. This is no longer simply a technology race; it is taking on the character of an "infrastructure race."
TSMC's Resolve to Hold the "Oil" of the AI Era
This report of "three additional fabs" serves as evidence that TSMC views the AI revolution not as a temporary boom, but as a long-term turning point in industrial structure.
An investment of NT$900 billion is a scale that would test the risk tolerance of any single company, but with demand from major customers like Apple and NVIDIA effectively guaranteed, it likely represents a "bet with strong odds" for TSMC.
While rival companies remain stuck in place, TSMC has further accelerated into territory that competitors simply cannot catch up to, backed by overwhelming investment scale and technological prowess. If the new fabs in the STSP special zone come to fruition, southern Taiwan's semiconductor cluster will further cement its status as the world's largest and most advanced AI chip manufacturing hub.
We are now witnessing an era in which silicon (semiconductors) has become a strategic resource even more critical than oil, and where the capacity to supply it determines the competitiveness of companies—and, by extension, of nations.
Sources