The world's largest semiconductor foundry is navigating a paradox that defines its current market position: demand is so robust that it must refuse customers, yet it is simultaneously taking bold measures to ensure long-term supply resilience. TSMC's stock price has surged nearly 125% over the past year, with a weekly close at €407, just slightly below its all-time high of €414. This rally reflects both the AI boom and investor confidence in the company's ability to manage its own success—confidence that was briefly tested last week when a Nikkei Asia report triggered a roughly 3.5% stock decline, followed by a swift rebound.

The report revealed that BYD, AMD, and Google are exploring the possibility of having Samsung fabricate future chips. BYD is targeting autonomous driving assist processors, AMD is considering sourcing certain CPUs from Samsung beginning in 2028, and Google is evaluating options for its Axion processors and TPU components. The reason is straightforward: TSMC's advanced process capacity is fully booked by customers including NVIDIA, Apple, AMD, and Broadcom, leaving virtually no room for new major clients. As a Chinese automotive chip manager interviewed by Nikkei put it, TSMC prioritizes advanced manufacturing investments because the margins are higher and capacity is tight. Samsung's available capacity, despite lower yields, is becoming increasingly attractive.

Yet TSMC is not standing still. The company recently signed a historic ten-year agreement with Amkor Technology to establish a seamless chip supply chain on U.S. soil. Under this deal, TSMC will supply raw silicon wafers from its Arizona fab, while Amkor will handle complex packaging and final testing—tasks that previously required shipping chips back to Asia. This arrangement is a game-changer for regulated buyers such as the U.S. military, who now have domestic sourcing for the entire chip production cycle.

Packaging bottlenecks are central to both the semiconductor industry narrative and market discussion. TSMC Chairman Wei Che-hua has repeatedly warned of structural capacity constraints across the entire supply chain—power, chips, and equipment. The partnership with Amkor represents a direct response to this shift, but the company is also aggressively expanding its own advanced packaging capacity. Should investors sell immediately? Or is TSMC worth buying?

Currently, TSMC's CoWoS (wafer-level chip-scale packaging) technology faces roughly a 20% supply-demand mismatch, critical for AI accelerators. This gap is expected to narrow to approximately 10% by the end of 2026. TSMC's monthly CoWoS capacity is projected to reach 120,000 to 140,000 wafers, with partner suppliers contributing an additional 50,000 to 60,000 wafers, bringing total industry capacity close to 200,000 wafers per month.

Meanwhile, TSMC is advancing its next-generation packaging technology, CoPoS (panel-level chip-scale packaging). Since 2025, its subsidiary Wafer Works has been running a pilot production line, with material and equipment certification expected to complete in June 2026. NVIDIA's Feynman platform is a leading candidate for early adoption, though mass production is not expected until 2028 at the earliest.

On quality, TSMC maintains high standards. Its 5.5-tile CoWoS solution achieves yields exceeding 98%. By contrast, Intel's EMIB-T technology hovers around 90%, while Samsung's 2nm process sits in the mid-50% range. This performance gap explains why, even as customers explore alternatives, they remain reluctant to leave TSMC.

The market recognizes that refusing customers is not a sign of weakness. The stock's 50-day moving average stands at €342, approximately 19% below current prices—a technical indicator pointing to strong upside momentum. Analysts have formed a bullish consensus: all 17 experts covering the stock have issued buy ratings, with a consensus target price of $473.40. Is TSMC at an inflection point? This analysis reveals what investors need to know now.

TSMC's 2026 capital expenditure budget is $52–56 billion. Of this, 70–80% will be directed to advanced process nodes, approximately 10% to specialty process nodes, and the remainder to packaging, testing, and mask production. Advanced packaging accounted for roughly 8% of revenue in 2025; this share is expected to exceed 10% for the first time in 2026.

The long-term thesis rests on structural scarcity. AI infrastructure, high-performance computing, and automotive electronics are simultaneously pressuring capacity. TSMC's response—locking in U.S. packaging capacity via Amkor, expanding CoWoS, and investing in CoPoS—signals that management expects demand surges to persist through the next decade. For investors, the news of Samsung "poaching" customers is not a warning but confirmation that TSMC's order book is so full that even its own clients are forced to seek alternatives elsewhere.