--- title: "After Jensen Huang's visit to Taiwan and his request for \"production capacity,\" is Taiwan Semiconductor urgently adding expansion plans? It may raise capital expenditures for 2026" description: "Morgan Stanley expects that in response to strong demand from major AI customers such as NVIDIA, Taiwan Semiconductor is considering increasing its 3-nanometer capacity target for 2026 by an additiona" type: "news" locale: "en" url: "https://longbridge.com/en/news/265831663.md" published_at: "2025-11-14T02:22:55.000Z" --- # After Jensen Huang's visit to Taiwan and his request for "production capacity," is Taiwan Semiconductor urgently adding expansion plans? It may raise capital expenditures for 2026 > Morgan Stanley expects that in response to strong demand from major AI customers such as NVIDIA, Taiwan Semiconductor is considering increasing its 3-nanometer capacity target for 2026 by an additional 20,000 wafers per month on top of the original plan, bringing total capacity to 160,000-170,000 wafers per month. This move could significantly boost its capital expenditure for 2026 to $48-50 billion, far exceeding market expectations. This is also seen by the market as a positive catalyst for the global semiconductor equipment industry "Without TSMC, there would be no NVIDIA"! Jensen Huang went to the door to "request goods", and TSMC is reported to be considering expanding its capacity. On November 14th, according to news from Chasing Wind Trading Platform, Morgan Stanley's latest research report indicates that TSMC may plan to add an additional 20,000 wafers per month of 3nm wafer capacity beyond the current expectations. The firm initially estimated TSMC's 3nm capacity in 2026 to be around 140,000 to 150,000 wafers per month, but the latest information suggests that this figure may be raised to 160,000 to 170,000 wafers per month. This move closely follows NVIDIA CEO Jensen Huang's recent high-profile visit to the company. During the visit, Huang publicly stated that NVIDIA's business is "very strong" and has requested TSMC to provide more chip supplies. If this expansion plan is implemented, it will have a direct impact on TSMC's capital expenditures. It is estimated that the additional capacity will require approximately $5 billion to $7 billion in extra capital expenditures, potentially raising TSMC's total capital expenditures for 2026 from the current expectation of about $43 billion to a range of $48 billion to $50 billion. This potential increase in spending is seen by the market as a positive catalyst for the global semiconductor equipment industry. This move not only reflects the continued strong demand for AI but also further confirms that the main contradiction in the current AI supply chain has shifted from backend packaging to frontend manufacturing. The report clearly states that CoWoS advanced packaging is no longer the biggest constraint, and the real shortage lies in TSMC's frontend wafer capacity and key materials such as ABF substrates. ## **3nm capacity is in urgent demand, TSMC may shift old production lines for expansion** As customers like NVIDIA, AMD, and Alchip rush to place orders, TSMC's 3nm capacity is becoming increasingly tight. The report points out that the industry previously believed that TSMC wanted to maintain full utilization of its advanced process capacity, thus taking a cautious approach to expansion. However, after major AI customers like NVIDIA made requests, its 3nm capacity plans may have changed. According to TSMC management's statements during the earnings call, all newly built cleanroom space in Taiwan will be used for the expansion of 2nm processes. This means that the expansion of 3nm will need to take place within existing wafer fabs. To address the space issue, Morgan Stanley's latest survey indicates that TSMC may consider moving its 22/28nm equipment from the Fab15 wafer fab and reserving this equipment for future use in European factories, thereby freeing up valuable cleanroom space for the new 3nm production line. According to the firm's predictive model, **this adjustment will raise TSMC's total 3nm capacity in 2026 from the previously estimated 140,000 to 150,000 wafers per month to 160,000 to 170,000 wafers per month.** **** ## **Capital expenditures may be significantly increased, benefiting equipment manufacturers** Behind capacity expansion is a massive capital investment. Analyst Charlie Chan estimated in a report that with a capital expenditure of $300 million required for every 1,000 3-nanometer wafers, an additional capacity of 20,000 wafers per month would correspond to an investment of $5 billion to $7 billion, potentially pushing TSMC's total capital expenditure for 2026 to a range of $48 billion to $50 billion. This additional expenditure will significantly raise TSMC's capital expenditure forecast for 2026. The report believes that this potential increase in capital expenditure is positive for the global semiconductor equipment industry sentiment. The firm also reaffirmed its "overweight" rating on industry companies such as TSMC, King Yuan Electronics, and ASE Technology. Changes in TSMC's capital expenditures are a key barometer for observing the semiconductor industry's prosperity. An increase in capital expenditure typically indicates that the company has strong confidence in future market demand, which will directly benefit upstream equipment suppliers. ## **CoWoS is no longer a bottleneck, front-end wafers become key** For some time, the market has generally viewed CoWoS advanced packaging technology as the main bottleneck for AI chip supply. However, the report emphasizes that this view may be outdated. The report analyzes that although major tech companies have announced large-scale data center power deployment plans, when translating this into demand for CoWoS, calculations show that TSMC and other manufacturers' CoWoS capacity should be sufficient to meet the demand. The report points out that the real bottleneck lies in front-end wafer manufacturing capacity, as well as the supply of other materials such as ABF substrates and T-Glass (glass substrates). This judgment has also been indirectly confirmed by the industry chain. According to a Reuters report, ASIC design service company Alchip has reiterated the dilemma of 3-nanometer wafer shortages. This indicates that even if back-end packaging capacity is sufficient, if front-end wafer supply is inadequate, the final output of AI chips will still be limited. ## **AI demand remains strong, customers like NVIDIA scramble for capacity** The fundamental driving force behind TSMC's consideration of urgent capacity expansion stems from the insatiable "computing power hunger" of AI giants led by NVIDIA. Jensen Huang candidly stated during his visit to Taiwan, "Without TSMC, there would be no NVIDIA today," clearly revealing the deep dependence of its growth on TSMC's capacity. According to the report, in the composition of customer demand for TSMC's 3-nanometer process, NVIDIA and another U.S. customer are the largest sources of demand, and it is expected that they will dominate from 2025 onwards. Additionally, the report predicts that by 2025, AI-related revenue will account for 25% of TSMC's total revenue. From Tesla's 3-nanometer AI5 chip design to major cloud service providers' ongoing investments in AI servers, all signs point to one conclusion: demand for the most advanced process chips will continue to grow strongly in the foreseeable future. TSMC's consideration of capacity expansion is a direct response to this market trend and indicates that the global semiconductor industry's capital competition will continue The above wonderful content comes from Chasing Wind Trading Platform. For more detailed interpretations, including real-time analysis and frontline research, please join the【 Chasing Wind Trading Platform ▪ Annual Membership】. ### Related Stocks - [TSM.US - Taiwan Semiconductor](https://longbridge.com/en/quote/TSM.US.md) - [SOXL.US - Direxion Semicon Bull 3X](https://longbridge.com/en/quote/SOXL.US.md) ## Related News & Research | Title | Description | URL | |-------|-------------|-----| | 反超三星、利润率盖过台积电：SK 海力士是如何问鼎 AI 存储领域 “隐形霸主” 的？ | SK 海力士过去 12 个月市值飙升 340%，已从债权人控制的 “僵尸公司” 逆袭为 AI 产业链定价者。其凭借 HBM 先发优势，在 AI 需求爆发的浪潮下，深度绑定英伟达与微软，拿下全球过半 HBM 份额；市占率超越三星，利润率超越台 | [Link](https://longbridge.com/en/news/275576275.md) | | Stratechery 创始人深度访谈：预警 2029 年大规模 “芯片荒”，SaaS 模式将终结，广告才是 AI 终极商业闭环 | Ben Thompson 警告，台积电保守扩产或致 2029 年全球芯片短缺，巨头需分担建厂风险。在对巨头点评中，他认为 Meta 执行力最佳；谷歌混乱但有韧性；亚马逊自研芯片策略面临风险。展望未来，他指出在数字世界高度发达后，“现场” 体 | [Link](https://longbridge.com/en/news/276001569.md) | | Meta 加码英伟达：未来数年部署数百万颗芯片，首次采用 Grace CPU | 根据周二发布的声明，Meta 承诺将使用更多来自英伟达的 AI 处理器和网络设备。同时，Meta 还将首次在其独立计算机的核心部件采用英伟达的 Grace CPU。此次部署将涵盖基于英伟达当前 Blackwell 架构，以及即将推出的 Ve | [Link](https://longbridge.com/en/news/276177748.md) | | 闪迪启动 30.8 亿美元二级市场股票发行 | 闪迪于 2 月 17 日宣布启动 30.8 亿美元的二级市场股票发行，涉及现有股东出售股份。西部数据将通过股权方式处置对闪迪的债权，具体债务规模和交易细节尚未披露。投资需谨慎，本文不构成个人投资建议。 | [Link](https://longbridge.com/en/news/276180619.md) | | 段永平试水 “AI 交易”：卖苹果加仓英伟达，“新入” CoreWeave、Credo 和 Tempus | CoreWeave 专门搭建高性能 GPU 集群，把算力租给 AI 公司和企业客户；如果说 GPU 是心脏，Credo 提供的高速互联芯片和光模块就是血管，跟 AI 服务器迭代高度绑定；Tempus 则致力于将 AI 应用于精准医疗，尤其是 | [Link](https://longbridge.com/en/news/276189117.md) | --- > **Disclaimer**: This article is for reference only and does not constitute any investment advice.