Under the AI wave, chip demand is exploding. As TSMC's production capacity is fully booked until 2027, major clients like Apple and NVIDIA among the "Big Six" are forced to consider shifting some orders to Samsung and Intel. According to Zhui Feng Trading Desk, a latest research report from Deutsche Bank reveals that the global semiconductor foundry giant Taiwan Semiconductor Manufacturing (TSMC) is facing a "happy problem." Due to extreme shortages in 3nm process capacity, with orders filling the entire schedule for 2026 and extending into 2027, TSMC has been compelled to significantly raise its capital expenditure plans. This production bottleneck is reshaping the market landscape, forcing top-tier clients who previously relied on TSMC to seriously consider Samsung and Intel as alternative solutions. Analysts including Robert Sanders at Deutsche Bank pointed out in the report that TSMC's disclosed capital expenditure guidance for 2026 is as high as $52 to $56 billion, a figure substantially above the bank's expectation of $50 billion and the market consensus of $46 billion. Analysts believe this effectively amounts to TSMC admitting that its 2025 capital expenditure plan was "too conservative" in the face of soaring AI demand. The current situation is no longer simply an overflow from CoWoS advanced packaging capacity constraints, but a severe supply shortage at the core wafer fabrication level—especially for the 3nm process. This supply-demand imbalance is creating direct spillover effects in the market. The report emphasizes that, despite Samsung and Intel having a "chequered record" in foundry services, the "Big Six" clients—Apple, NVIDIA, AMD, Broadcom, Qualcomm, and MediaTek—currently have little choice but to seek alternative production capacity. As a result, TSMC's market share in advanced process node foundry is expected to decline from 95% to 90%. With capital expenditure surging to $56 billion, securing 3nm capacity has become exceptionally difficult. The most striking figure in TSMC's Q4 financial report was its capital expenditure guidance for 2026. The scale of $52 to $56 billion indicates the company is going all out to address the capacity crisis. Deutsche Bank noted that, considering TSMC's full-year 2026 capacity is likely already fully booked, with order backlogs stretching deep into 2027, this surge in spending is not surprising. Previously, the initial bottleneck following the AI demand explosion was mainly concentrated in the CoWoS packaging segment, leading to some orders spilling over to vendors like ASE and Amkor. However, the current situation is more severe: demand is far outstripping supply, particularly for the 3nm process. Data from the report shows that even though TSMC plans to increase its 3nm capacity to 190,000 wafers per month (190k wpm) by the end of ̈2026, it still cannot satisfy client demand. Deutsche Bank mentioned in the report: "While clients wanting more capacity is normal under TSMC's high utilization rates, the extent seen now is unprecedented."
The extreme capacity crunch is forcing the "Big Six" clients to diversify, with Samsung's Taylor fab potentially becoming the primary alternative. The severe capacity shortage is compelling TSMC to adopt more aggressive strategies. Reportedly, TSMC is delaying the start of new 3nm development projects, instead guiding clients to shift product plans for 2027/28 mass production more towards the 2nm GAA process. Concurrently, TSMC is also using pricing strategies to "nudge" clients towards this transition, evident from its strong Q1 2026 guidance. This suboptimal situation is forcing the world's six largest fabless chip companies—Apple, NVIDIA, AMD, Broadcom, Qualcomm, and MediaTek—to turn their attention to competitors. Deutsche Bank analysis suggests that in this capacity scramble, Samsung's fab in Taylor, Texas (SF2P) might be favored over Intel. The report states: "For customers seeking alternative supply, Samsung's Taylor fab is more likely to be the first port of call." Specifically, Qualcomm and AMD are most likely to consider Samsung; while based on previous discussions, Apple and Broadcom are reportedly evaluating Intel. However, analysts added that although Intel has potential with its 14A process story, it "still has a lot of work to do." AI drives upward revision of long-term growth expectations, with gross margin outlook raised to 56%. Despite short-term challenges in capacity allocation, the long-term growth红利 driven by AI is considered highly certain. TSMC confirmed that it expects AI-related growth to accelerate, revising its compound annual growth rate (CAGR) expectation for 2024-2029 from the mid-40% range previously to the mid-to-high-50% range. Driven by this, TSMC has raised its long-term overall company growth forecast to a CAGR of 25%, and its long-term gross margin target has been correspondingly raised to 56%. Deutsche Bank believes that, considering the latest developments in AI and TSMC's pricing actions, these upward revisions are "not surprising." Regarding overseas capacity expansion, the report mentions that TSMC's Arizona fab is currently in a low-volume phase (Phase 1: 4nm, 20k wpm capacity), with the first 3nm revenue from Phase 2 expected by the end of this year. Although overseas fab construction brings a 2-4% gross margin dilution and faces challenges related to talent, infrastructure, and yield, market focus remains primarily on its core profitability.
Based on improved profitability from higher wafer pricing, Deutsche Bank has raised its target price for TSMC (2330.TW) by 10% to NT$2,200. This target price implies a price-to-earnings ratio of 20x based on forecast 2027 EPS, aligning with peer levels. Analyst Robert Sanders stated that this valuation reflects TSMC's solid positioning and strong growth rate through 2028. However, the report also highlights potential risks, including geopolitical risks and potential resistance from Intel's attempts to bring production in-house. Furthermore, a sudden slowdown in AI-related spending or poor execution in ramping up CoWoS capacity would also pose risks.
Comments