This persistent shortfall is a structural feature of the AI transition. Wei noted that demand is spreading beyond data centers into personal computers, smartphones, automobiles, and IoT devices, which he argued points to sustained, broad-based growth . Despite rising component costs and geopolitical uncertainties, Wei said customer behavior hasn’t changed and that clients remain “positive on the outlook for artificial intelligence”
.
When asked directly whether TSMC would raise prices, Wei was unusually candid. “I would like to do that,” he said, acknowledging the need to maintain profitability while investing tens of billions of dollars in new capacity . His remark drew attention because it signaled a shift in posture from a supplier that has historically been cautious about antagonizing its largest customers, which include Apple and Nvidia.
However, Wei immediately emphasized restraint. He pledged that TSMC would not adopt the pricing practices of the memory-chip industry, where producers implement sharp increases during supply shortages. Instead, the company will refrain from sudden price hikes to preserve stable, long-term customer relationships . The underlying message was that price increases are likely, but they will be gradual and predictable rather than opportunistic.
The pricing discussion was set against the backdrop of TSMC’s best-ever financial year. Wei reported that 2025 delivered record revenue and earnings per share, both reaching all-time highs . The company’s stock price surged from NT$950 per share at the previous year’s meeting to NT$2,425 on June 3, 2026—a gain of more than 150%
.
Shareholders also benefited from TSMC’s performance. The company announced a cash dividend of at least NT$24 per share for the year, representing growth of more than 30% compared to the prior year . These numbers reflect the explosive revenue growth driven by AI-related chips from customers like Nvidia and AMD, which have been competing fiercely for TSMC’s most advanced process nodes.
Wei gave shareholders an aggressive growth forecast for 2026. TSMC expects its dollar-denominated revenue to grow more than 30% for the full year, driven entirely by the AI demand cycle .
To meet that demand, TSMC plans to increase capital expenditure to approximately $56 billion in 2026, a roughly 37% increase from the $40.9 billion it spent in 2025 . This record investment scale is aimed at expanding capacity for 3nm and more advanced process technologies, including the upcoming A16 node scheduled for the second half of 2026
. The capex number exceeded market expectations and underlined the company’s conviction that the AI boom is durable, not a temporary spike.
A recurring shareholder concern was whether TSMC can maintain its dominance as rivals mobilize. Wei addressed these worries directly and dismissively.
He said TSMC is “not afraid” of competition from mainland Chinese foundries or advances by Huawei Technologies, stating that competition has been a constant throughout the company’s four-decade history . He brushed off specific threats from Intel’s 18A process, Samsung Foundry, and even Elon Musk’s rumored TeraFab facility, wishing Musk “well” in a tone that suggested he sees no credible near-term threat to TSMC’s technological leadership or manufacturing efficiency
.
“TSMC has never lacked competitors,” Wei said. His formula for staying ahead is not secrecy but relentless execution and commitment to research and development . While TSMC is expanding its production footprint across Taiwan, Japan, Germany, and the United States, Wei’s tone underscored that Taiwan’s ecosystem of talent, supply chains, and infrastructure remains the company’s core competitive advantage and an irreplaceable link in the global AI supply chain.
Wei’s 2026 shareholder address reinforced a straightforward thesis: AI demand is structurally larger than current supply, TSMC intends to capture its fair share of margin through gradual price increases, and despite the best efforts of global competitors, the company’s Taiwanese foundation gives it a multi-year lead that is difficult to replicate.
Comments
0 comments