TSMC delivered a record-breaking quarterly earnings report, demonstrating that demand from chip manufacturers shows no signs of slowing down—TSMC’s production capacity is already struggling to keep up with market demand.
The AI boom has created overwhelming demand for production capacity at this world-leading chip foundry, potentially constraining the entire industry’s ability to expand. Although TSMC disclosed its 12-inch wafer equivalent capacity last year, customers are seeking even more.
Mizuho Securities analyst Jordan Klein said: "The competition is extremely fierce—it's a battle to gain greater influence and secure larger supply allocations."
The vast majority of the world’s most advanced semiconductors—NVIDIA, AMD, and Broadcom GPUs and CPUs required to run AI workloads—are manufactured by TSMC. Today, an increasing number of tech companies—including Apple, Google, Amazon, Microsoft, Meta, and OpenAI—also rely on TSMC to produce their own custom AI accelerator chips.
TSMC’s North America CEO, Sajiv Dalal, revealed that he receives at least 10 calls daily inquiring about wafer capacity. On Thursday, TSMC reported its sixth consecutive quarter of strong financial performance; driven by robust demand for advanced-process chips, the company expects fourth-quarter revenue to surge 58% year-over-year.
Klein stated that if computing power demand continues to rise at the current pace over the next 12 to 18 months, TSMC’s capacity constraints could disrupt massive infrastructure investment plans across the industry and the “aggressive revenue growth targets” set by AI companies. U.S. hyperscale cloud providers have already combined for capital expenditures exceeding $700 billion this year, and analysts expect this figure to continue rising by 2027.
The development of AI is also constrained by a global shortage of memory chips. The logic chips manufactured by TSMC have higher process complexity and higher production costs. As a result, industry leaders such as NVIDIA and AMD have been engaging in closed-door negotiations with TSMC years in advance, signing long-term contracts and customizing exclusive collaboration plans. Both NVIDIA’s CEO Jensen Huang and AMD’s CEO Lisa Su have visited Taiwan multiple times this year.
Sid Sheh, CEO of AI chip startup D-Matrix, has had a partnership with TSMC for over 20 years, with the company relying on TSMC to manufacture its processors. Sheh noted that for many of TSMC’s customers, business ultimately converges on Kai Chang, TSMC’s Senior Vice President, who is the key decision-maker in allocating precious production capacity to chip manufacturers.
Xie Si said in an interview: “He is probably the most popular person in the world today.” Jerry Kuban, Vice President of Manufacturing at D-Matrix, has been working with TSMC since 1992 and travels to Taiwan three or four times a year.
Xie Si said, “At that time, NVIDIA hadn’t even been founded. We brought him on board specifically for his connections at TSMC.” Kuban joined D-Matrix in 2024.
Invest in startups
In 2025, TSMC's top ten customers accounted for a significant portion of the company's revenue, with the first and second largest customers contributing 19% and 17% of revenue, respectively.
This means startups must compete with much larger customers for production capacity. Xie Si said that TSMC’s clients are highly selective, but there is also motivation to support promising emerging companies.
When NVIDIA and Broadcom were still small companies, TSMC strongly supported them; now, both are TSMC’s largest customers. TSMC hopes history will repeat itself.
TSMC is accelerating the expansion of its production capacity by building new factories in Taiwan, Japan, and Arizona, marking the first time advanced-node chips are being produced in the United States. In July of this year, TSMC announced that the number of planned wafer fabs in Arizona has increased from six to “several more,” with the total committed investment reaching $265 billion. Meanwhile, reports indicate that TSMC is also evaluating the possibility of constructing a full-scale wafer fab in Texas.
However, all of this new capacity will take several years to come online. Analysts say that during this transition period, TSMC is still purchasing ASML’s extreme ultraviolet (EUV) lithography machines to install in its existing factories, thereby increasing capacity. ASML stated in its Wednesday earnings report that customer equipment orders are now booked through 2030.
TSMC is not the only option. Capacity constraints have created opportunities for two other leading advanced chip manufacturers, Samsung and Intel, to attract customers.
Dan Hutcheson, an analyst at TechInsights, said: "Companies all want a second source of supply to balance TSMC and constrain its pricing."
Intel is the only advanced logic chip foundry based in the United States. Although its foundry business has faced years of challenges and setbacks, Hutchinson said that major chip companies are now conducting tapeouts at Intel.
The semiconductor industry association SEMI predicts that by 2029, 102 new semiconductor wafer fabs will come online globally, including 15 in the United States and 78 in Asia. China is the primary growth driver, with domestic chip production capacity expected to nearly triple by 2030. To catch up, U.S. fab investments are projected to more than double by 2029.
Klein's assessment: The ongoing establishment of the Arizona facility, combined with intense competition for wafer production capacity, positions TSMC as one of the companies with the most advantageous global strategic positioning.