報告_GS_台積電2330_20260910

PDF 原檔：報告_GS_台積電2330_20260910_original.pdf

原始內容

Goldman

For the exclusive use of [讀者資訊已遮蔽]

Research

Explore >

TSMC (2330.TW): Communacopia + Technology 2026 - Key Takeaways

Evelyn Yu

+886(2)2730-4187 | evelyn.yu@gs.com Goldman Sachs (Asia) L.L.C., Taipei Branch

James Schneider, Ph.D.

+1(212)357-2929 | jim.schneider@gs.com Goldman Sachs & Co. LLC

Anmol Makkar

+1(212)357-1366 | anmol.makkar@gs.com Goldman Sachs & Co. LLC

Luya You

+1(212)902-5297 | luya.you@gs.com Goldman Sachs & Co. LLC

Khalil Fenina

+1(212)357-6392 | khalil.fenina@gs.com Goldman Sachs & Co. LLC

Marshall Wong

+1(646)446-0222 | marshall.wong@gs.com Goldman Sachs & Co. LLC

Presenters: Wendell Huang, SVP & CFO

Bottom Line: We have three key takeaways: (1) TSMC continues to view AI as a multi-year megatrend that remains at a very early stage, with lower compute costs expected to accelerate AI adoption and ultimately drive greater demand for advanced silicon; (2) execution in Arizona continues to progress well, with Phase 1 achieving yields comparable to Taiwan and already generating pro fi ts, while TSMC expects approximately 30% of its N2-and-beyond capacity to be located outside Taiwan in fi ve years, primarily in Arizona; and (3) despite dilution from the N2 ramp and overseas expansion, management remains comfortable achieving its long-term gross margin target of 56% and above, supported by disciplined capacity planning, high utilization, productivity improvements, cost reductions, and pricing that re fl ects TSMC's value.

Key Takeaways

• AI remains in the early stages of a multi-year megatrend: TSMC continues to n see increasing adoption of AI. Management also highlighted the shift from generative AI toward agentic AI, which should broaden compute demand beyond accelerators to CPUs . TSMC remains con fi dent in the durability of the AI cycle based on discussions not only with its direct customers but also with its customers' customers. Recent visits of US CSPs further reinforced management's con fi dence, as cloud service providers appeared well prepared in areas including power and broader infrastructure deployment.
• Lower compute costs should drive greater consumption and silicon demand: n Management believes open-source models and continued improvements in computing e ffi ciency will reduce token costs, leading to greater AI consumption and broader adoption, supporting incremental demand for advanced silicon . TSMC has already incorporated this dynamic into its capacity planning and continues to expand aggressively to support what it views as

Goldman Sachs does and seeks to do business with companies covered in its research reports. As a result, investors should be aware that the fi rm may have a con fl ict of interest that could a ff ect the objectivity of this report. Investors should consider this report as only a single factor in making their investment decision. For Reg AC certi fi cation and other important disclosures, see the Disclosure Appendix, or go to www.gs.com/research/hedge.html. Analysts employed by non-US a ffi liates are not registered/quali fi ed as research analysts with FINRA in the U.S.

e92c7a75ab8b4efbba794e6b187208c8

For the exclusive use of [讀者資訊已遮蔽]

structural, multi-year demand growth.

• Disciplined demand validation remains critical amid aggressive expansion: n TSMC uses a combination of top-down and bottom-up analysis to validate demand. Its top-down assessment considers macroeconomic conditions, semiconductor demand, fabless/system companies, the foundry industry, etc, while its bottom-up approach evaluates demand across di ff erent segments and incorporates feedback from customers and customers' customers. Management acknowledged that semiconductors will remain cyclical, but believes correctly identifying the underlying megatrend is more important than attempting to forecast every short-term fl uctuation.
• Arizona execution is demonstrating that TSMC's manufacturing model can n travel: TSMC reiterated that overseas expansion is based on customer demand for geographic fl exibility and an appropriate level of government support. Arizona Phase 1, using N4 technology, is in operation, with yields comparable to Taiwan, satisfactory quality and reliability, and positive pro fi tability . Management views this as an important milestone because it demonstrates internally and to customers that TSMC can produce wafers in the US with the same quality as in Taiwan. Equipment move-in for Phase 2, which will use N3 technology, is expected to begin soon; construction of Phase 3 has started; and preparation work for Phase 4 and the fi rst US advanced-packaging facility is underway.
• Arizona has substantial room for further expansion: In addition to its original n Arizona development, TSMC has acquired another parcel of land capable of accommodating an additional fi ve to six fabs, supporting its recently announced incremental US$100bn investment . Based on its current plan, TSMC expects approximately 30% of its N2-and-beyond capacity to be located outside Taiwan in fi ve years, primarily in Arizona. However, the most advanced technologies will continue to ramp fi rst in Taiwan because the early-stage technology transfer requires close collaboration and frequent movement of engineers among R&D and manufacturing sites. Overseas deployment will generally follow after the technology has stabilized in Taiwan.
• Customers increasingly value and are willing to pay for geographic fl exibility: n Management noted that customers initially requested only a modest amount of geographically diversi fi ed supply, but their requirements have increased as TSMC has demonstrated successful execution in Arizona. TSMC believes this geographic fl exibility represents incremental value and intends to re fl ect that value in its pricing, with customers increasingly willing to recognize the associated premium.
• Technology, manufacturing excellence, capacity, and trust underpin TSMC's n competitive moat: Management believes customers select a foundry partner based on access to the best technology, the ability to ramp rapidly into high-volume manufacturing with attractive economics, su ffi cient capacity, and con fi dence that the foundry will not compete with its customers. TSMC is con fi dent in maintaining this lead by continuing to advance its roadmap. Management also highlighted its record of rapidly ramping new technologies, the scale of its wafer and advanced-packaging investments, and its consistent delivery against customer

e92c7a75ab8b4efbba794e6b187208c8

For the exclusive use of [讀者資訊已遮蔽]

• commitments as key elements of customer trust that cannot be replicated through CapEx or engineering recruitment alone.
• Pricing remains strategic rather than opportunistic: TSMC does not intend to raise n prices simply because near-term utilization is high, nor reduce prices simply because utilization temporarily weakens. Instead, the company's approach is to price sustainably and consistently based on the value it provides, while ensuring that customers can remain successful over the long term . Management noted that TSMC's structural pro fi tability has continued to improve, with its long-term gross margin target increasing from 50% fi ve years ago to 53% and now 56% and above.
• Leading-edge wafer and advanced-packaging capacity remain tight: n Management continues to build capacity wherever possible to meet customer demand. N3 demand remains particularly strong: unlike prior technology cycles, when demand for a node typically began declining two to three years after its introduction, TSMC is adding further N3 capacity and still does not expect supply to be su ffi cient . The company is also expanding advanced-packaging capacity to address strong accelerator demand. Outside TSMC, management identi fi ed power and memory as potential bottlenecks cited by other parts of the AI supply chain, while emphasizing that foundry capacity itself remains di ffi cult to expand because fabs require two to three years to build and another one to two years to ramp.
• TSMC's disclosed AI revenue understates its broader exposure: Management n reiterated that its disclosed AI revenue de fi nition is relatively narrow and primarily captures GPUs, custom AI accelerators and memory base die. It excludes CPUs and networking products because TSMC cannot reliably determine whether each chip is ultimately deployed in an AI system. As agentic AI drives incremental demand across x86, Arm, and RISC-V CPUs, as well as networking and other supporting silicon, TSMC's total AI-related exposure is therefore broader than its reported AI-accelerator revenue. The company intends to retain its existing de fi nition until it can more accurately distinguish the end use of these products.
• No visible barrier to continued transistor scaling: TSMC is currently ramping N2, n with A14 expected approximately two years later, while development of the subsequent node is already underway. Management does not currently see a technical issue preventing delivery of the next node or continued advancement beyond the existing roadmap. TSMC views close collaboration with customers as a key foundry advantage, allowing the company to develop each technology against speci fi c customer performance, power, density, and economic requirements.
• Cross-node capacity optimization is an underappreciated advantage: TSMC n highlighted its ability to generate additional output through productivity improvements and to convert tools and capacity across technology nodes. For example, the company can use portions of its N5 capacity to support N3 demand and can also optimize capacity across less-adjacent nodes where technically feasible. Management views this cross-node fl exibility as a distinctive element of TSMC's manufacturing excellence, allowing the company to respond more quickly to demand changes while reducing the risk of capacity obsolescence .
• Advanced packaging is strategic primarily because it enables wafer growth: n

e92c7a75ab8b4efbba794e6b187208c8

For the exclusive use of [讀者資訊已遮蔽]

Valuation/Risks

CoWoS capacity remains tight, and TSMC already started outsourcing selected packaging processes to OSAT partners to support additional growth. Management emphasized that advanced packaging is intended primarily to ensure and enable wafer demand rather than serve as a standalone pro fi t pool . TSMC is therefore developing multiple packaging architectures, including panel-level packaging, etc, while remaining open to customers using alternative packaging providers, as either approach ultimately supports demand for TSMC wafers.

• Advanced-packaging execution remains a key di ff erentiator: TSMC noted that its n current advanced-packaging capability supports approximately 5.5x reticle size, with a roadmap to reach approximately 14x reticle sizes within the next several years. While competing roadmaps may also appear attractive, management believes execution and the ability to deliver in high-volume manufacturing will ultimately determine competitive outcomes. TSMC does not believe competitors can use advanced-packaging wins to meaningfully displace its leading-edge wafer business, as foundry and packaging remain fundamentally di ff erent businesses .
• A uni fi ed TSMC culture supports overseas manufacturing execution: TSMC's n objective is not to create separate Taiwan and US cultures, but to establish a consistent TSMC culture centered on integrity, commitment, innovation, and customer trust (ICIC). Locally hired Arizona employees spend six to eighteen months training and working in Taiwan, while experienced Taiwan-based employees are assigned to Arizona to work alongside local teams. Over time, the number of Taiwan assignees should decline as locally trained employees assume greater responsibility. TSMC is applying this approach across its overseas manufacturing sites.
• Management remains comfortable with its 56%-and-above long-term gross n margin target: TSMC reiterated its 6 major drivers of pro fi tability: new-technology ramp, pricing, cost, product mix, utilization, and FX. The N2 ramp is expected to dilute 2026 GM by approximately 2-3ppts, but its pro fi tability should improve over time toward the corporate average; management also noted that N2's structural pro fi tability is better than N3 at a comparable stage. Overseas expansion is expected to dilute GM by 2-3ppts during the early portion of the current fi ve-year period, then widening to 3-ppts later as more overseas fabs begin production. Each new fab will initially create dilution before improving along its learning curve. Taking these puts and takes together, management remains very comfortable that disciplined capacity planning, healthy utilization, cost reduction, productivity gains, and pricing that re fl ects TSMC's value will enable the company to deliver for shareholders and achieve its long-term gross margin target 'of 56% and higher' with focus on the 'higher' side .

TSMC (Buy) : Our 12-month price target is NT$3,100, which is derived by applying a target P/E multiple of 22x to our 2027E EPS. Our 22x target P/E is benchmarked against 1.0stdv above its 5-year trading average. For the ADR (TSM), we have a 12-month price target of US$620, based on a USD/TWD rate of 30.0 and an ADR premium of 20%. Key e92c7a75ab8b4efbba794e6b187208c8

For the exclusive use of [讀者資訊已遮蔽]

downside risks to our views: (1) further deterioration in end-demand recovery impacting capacity utilization; (2) slower customer node migrations; (3) slowdown in AI investment resulting in lower long-term semiconductor content growth; (4) poor yields/execution resulting in worse-than-expected pro fi tability; (5) stronger competition resulting in ASP/pro fi tability erosion; and (6) unfavorable FX trend or higher-than-expected cost increase weighing on the margin outlook.

Source: Company data, Goldman Sachs Research estimates, FactSet. Price as of 9 Sep 2026 close.

e92c7a75ab8b4efbba794e6b187208c8

For the exclusive use of [讀者資訊已遮蔽]