TSMC Q1 Earnings: 2026 Capital Budget Raised to Meet AI Demand Surge
Foundry Giant Eyes Higher Margins as 3nm Reaches Full Depreciation in Late 2026, but Geopolitical Risks and Infrastructure Bottlenecks Loom
Seen as a barometer of future AI demand, Taiwan Semiconductor Manufacturing Co. (TSMC), which produces more than 90% of the advanced logic ICs used in AI servers, lifted its capital budget projection aligned with expectations for revenue growth.
TSMC has maintained capital expenditures at around 30% of revenue over the past three years. CFO Wendell Huang said the company’s 2026 capital budget is expected to come in at the high end of its US$52 billion to US$56 billion guidance range.
“With our strong technology leadership and differentiation, we are well positioned to capture the multi-year structure demand from the industry megatrends of 5G, AI, and high-performance computing (HPC),” said CFO Wendell Huang.
Chairman and CEO C. C. Wei explained that TSMC is stepping up 3 nanometer chip production to meet the multi-year robust demand of AI, smartphones, HPC, AI, including high-bandwidth memory (HBM) base dies, automotive and Internet of Things (IoT) customers.
“In Taiwan, we are adding a new 3-nanometer fab to our Gigafab cluster in Tainan Science Park, with volume production to start in 1H27,” said Wei, adding that TSMC is also converting 5nm tools in Taiwan to expand 3nm capacity. Across all regions, the company is improving manufacturing productivity and optimizing flexible capacity among its N7, N5, and N3 nodes to maximize wafer output and better serve customer demand.
Its global expansion is accelerating production of 3nm chips, too, with volume production at its second Arizona fab set for the second half of 2027, while in Japan, the company’s second fab will also adopt 3nm technology, with mass production planned for 2028.
However, despite all these efforts, supply is likely to remain tight over the next three years. “Building a fab takes 2-3 years, while ramping up production also takes 2-3 years,” said Wei, saying that Cloud Service Provider (CSP) customers’ demand remains robust. It is worth noting that whether CSPs design their own ASIC chips or purchase from NVIDIA or AMD, TSMC remains the only foundry capable of manufacturing them at scale and with the best yields.
The “A14” Leap and the Terafab Factor
TSMC also provided clarity on its next-generation A14 process. Utilizing nano-sheet transistor architecture, A14 aims for a 10–15% speed increase or 25–30% power reduction compared to the N2 node, alongside a 20% boost in chip density.
Despite these advancements, Wei warned that supply will remain tight. Addressing Elon Musk’s “Terafab” ambitions and competition from Intel, Wei was candid: “There are no shortcuts. The fundamental rules of the foundry game never change. It requires technology leadership, manufacturing excellence, and deep customer trust.” He hinted that while others may build factories, they lack the “Mission Impossible” service level that led Nvidia CEO Jensen Huang to famously compare TSMC’s agility to an “elephant that not only can dance but jump through hoops on fire for customers.”
A14 development remains on schedule, with strong customer interest from the smartphone and high-performance computing sectors. Volume production is planned for 2028.
Analysts also raised questions about two key challenges in the AI chip race: whether advanced packaging technologies such as SoIC and CoPoS can overcome larger die sizes, warpage, and thermal constraints in AI superchips, and whether rising competition from Samsung Foundry in AI processors, such as Groq’s LPU and Tesla’s AI4 and AI5, could threaten TSMC’s position. C.C. Wei said the company is already working with the customer on the next-generation products, and is confident in overcoming both technical and competitive challenges, citing its engineering expertise, deep customer collaboration, and continued technology leadership.
Optimizing Capacity Mix
In addition to building new fabs to increase 3-nanometer chip production, Wei disclosed that it is also planning to wind down mature-node chip production at FAB2, which is a 6-inch wafer fab, and FAB5, which is an 8-inch FAB, and use available space to optimize the support for leading-edge applications.
However, Wei insisted that TSMC’s strategy in mature nodes has not changed, but more mature-node capacities have shifted towards overseas production in Japan’s JASM for CMOS image sensor application and ESMC in Germany for automotive and industrial applications, while building high-yield for those specialized technologies.
“Our strategy is to optimize our capacity mix within mature nodes and focus on the higher value-added and strategic segment while ensuring we have the necessary capacity to support our customers’ growth,” said Wei.
Previously, TSMC CEO visited Japanese Prime Minister Sanae Takaichi for her support of TSMC’s plans to make the second fab an advanced-node chip production site, and it has been moving the mature-node equipment from Taiwan to Japan to help the Kumamoto Fab 1 become profitable. Wei’s talk about the mature node seemed to confirm the speculation. Since the equipment is fully depreciated, those two overseas fabs focusing on mature-node chips will be able to break even sooner than expected.
One analyst pointed out that since the 3-nanometer process went online in 2022, and that equipment is likely to be fully depreciated by 2027. Actually, TSMC’s profit margin of 3-nanometer processes is likely to exceed the company's average by the second half of 2026. “We don’t have a number to share with you, but after the full depreciation, as our previous experience, the gross profit margins are generally very high,” said Wendell Huang.
The Infrastructure Bottleneck: Power and Geopolitics
Unsurprisingly, TSMC’s first-quarter revenues, earnings, and profit margins all exceeded previous guidance and market estimates again.
While TSMC’s Q1 financials beat all estimates, the horizon is not without clouds. The company warned that Middle East conflicts could spike the cost of specialized chemicals and gases, potentially dampening consumer confidence and diluting earnings by 2–4% as overseas fabs ramp up.
Closer to home, energy stability remains a "high-wire act." As Taiwan phases out nuclear power, the pressure on the grid to support massive new fabs is immense. This mirrors a growing crisis in the U.S.; LinkedIn co-founder Reid Hoffman recently noted that the primary bottleneck for American AI expansion isn't just about TSMC’s capacity bottleneck or chip access, but a critical shortage of power transformers. “Transformer lead times have stretched to as long as 2029, delaying many planned data center projects, with only about 5 gigawatts of 16 gigawatts announced currently under construction”. With a heavy reliance on Chinese imports for electrical infrastructure, the "AI Revolution" faces a physical reality check that silicon alone cannot solve.