TSMC's $60-64B capex turns the AI supply constraint into a spending commitment

Taiwan Semiconductor Manufacturing Co. told the market it expects 2026 revenue to grow slightly more than 40%, and lifted its capital expenditure guidance to between $60 billion and $64 billion, a substantial increase on earlier estimates ^[1][2]. The revenue line is a forecast and can be walked back in a quarter; the capex line is concrete, tooling orders and cleanroom schedules, which makes the AI supply constraint a decision TSMC has already funded rather than one it is still modelling.

The near-term demand data supports the posture. July 2026 revenue came in at NT$467.58 billion, up 45% against the same month last year ^[3], which is running roughly five points ahead of the full-year growth guide ^[4]. TSMC holds about 73% of the foundry market in advanced nodes as of Q1 2026 ^[5], and the spending is driven by sustained AI investment from customers including Nvidia, which needs the most advanced nodes for its GPUs ^[6]. Chief executive C.C. Wei used June to emphasise the company's commitment to meeting customer demand during what he described as an ongoing AI boom ^[7]. That is a supplier telling buyers it will not be the bottleneck, which is a different message from a supplier telling shareholders margins are safe.

On the cost side, TSMC plans to begin raising chip prices by up to 10% from 2027 to offset rising materials and equipment expenses ^[8]. Note the sequencing: the heavy spending year is 2026 and the pricing offset does not begin until 2027 ^[9]. Arizona is part of the reason. Building leading-edge fabs on American soil is expensive, and those costs are a meaningful contributor to the larger capex budget, according to the source reporting ^[10]. Geographic diversification is being paid for out of the same envelope as the AI capacity.

Which brings you to the multiple. Shares are trading in the $400 to $430 range on a forward price-to-earnings ratio of around 21-22x, while the trailing multiple sat near 31x in mid-August 2026 ^[11][12]. The spread between the two is not a discount, it is an assumption: closing 31x down to 21-22x requires earnings to rise roughly 41% to 48% ^[13]. In other words, the forward multiple only looks moderate if the $60 billion to $64 billion converts into earnings more or less on schedule. Some analysts have flagged overvaluation risk on exactly that basis, arguing the elevated multiples already price the most optimistic growth scenarios ^[14]. SoftBank has reportedly been divesting much of its stake ^[15].

Three things to watch. First, whether monthly revenue keeps printing above the 40%-plus annual pace ^[3][1], because the guide now has a fixed cost base sitting behind it. Second, how the 2027 price increase lands: with 73% advanced-node share, customers either absorb the cost or pass it downstream, and they do not have many alternatives ^[5][16]. Third, the split of the capex between capacity that ships quickly and Arizona capacity that carries a structural cost premium ^[10] - the same dollar does not earn the same return in both places.