TSMC Sets Record in Q2 and Raises Guidance to Over 40% with AI Driving 66% of Revenue

Revenue of $40.2 billion grows 33.7% in dollars, gross margin reaches 67.7%, and capex increases to up to $64 billion. C.C. Wei discusses a new AI industry.
Taiwan Semiconductor Manufacturing Company reported consolidated revenue of $40.20 billion for the second quarter on Thursday, marking a 33.7% year-over-year increase and a 12.0% sequential rise. Net profit reached NT$ 706.56 billion, soaring 77.4% year-over-year, making it the highest quarterly result ever recorded by the Taiwanese manufacturer. The gross margin increased to 67.7%, while the operating margin rose to 60.3%, and the company raised its growth guidance for 2026 to "slightly above 40%" in dollars.
The underlying trend is clear: the demand for AI chips has transformed TSMC's revenue mix. The high-performance computing segment, which includes data center accelerators, accounted for 66% of quarterly revenue, with a 20% sequential increase in just one period. The 3-nanometer node represented 30% of wafer revenue, while the 5-nanometer node accounted for another 33%, indicating that orders are concentrated in the more expensive geometries of the lineup, where the Taiwanese firm has no commercial competitors in volume.
Wei Speaks of a "New Industry"
In the conference call, CEO C.C. Wei went further than usual in his public statements. "The megatrend of AI continues to drive the need for increasing computing power, which supports robust demand for cutting-edge silicon," he stated, adding, "we are witnessing a kind of new industry called the AI industry." According to Wei, the emergence of agent-based AI is expanding CPU consumption beyond accelerators, creating a favorable side effect for the company. "Regardless of the CPU approach adopted, whether x86, Arm, or RISC-V, almost all of them are TSMC customers."
Wei's comments on CPUs run counter to the narrative circulated in the first half of the year that the AI cycle would cannibalize traditional processing. For TSMC, the opposite scenario is materializing: each autonomous agent in production pulls GPUs for training and CPUs for inference and orchestration, and the company produces both.
Capex of $64 Billion and the Arizona Effect
The Taiwanese company raised its capital budget for 2026 to a range between $60 billion and $64 billion. The guidance for the third quarter is between $44.6 billion and $45.8 billion.
The ramifications of the capex extend beyond Taiwan. In the United States, the investment program announced earlier this year includes the construction of the Arizona complex, which will begin producing N3 and N2 nodes. For clients such as Nvidia, AMD, Broadcom, and Apple, the production schedule of Fab 21 is now a planning variable as relevant as the launch of the products themselves.
How the Queue Organizes
The point that consultancies and cloud architecture teams in global banks need to monitor is the allocation of capacity over the next twelve months. TSMC continues to sell production before it even exists, with wafer reservation contracts that already cover a significant portion of the N2. European sovereign cloud providers and Japanese robotics integrators, alongside customers like Fujitsu and Hitachi, share space on the same lines with American hyperscalers, and the allocation is not neutral: the queue is cash up front, and it prevails.
For the CIO of a bank in Frankfurt or an insurer in Tokyo, the practical message is not about the margin number but rather the explicit multiplier. While the market speculates about when data centers will stop purchasing accelerators at the current pace, the Taiwanese manufacturer is signaling the opposite: its customers are now signing contracts that will only produce silicon in 2028.
There is a legitimate counterpoint that needs to be noted. Analysts from Bernstein and Barclays have been warning that TSMC's elevated guidance depends on hyperscalers maintaining the current pace of orders, and IBM's 25% decline last week, due to clients reallocating budgets from software to AI hardware, shows that this reallocation has operational limits. The cautious thesis is that 2027 may reveal an inventory adjustment, with a visible effect on margins. Wei responded indirectly during the call, stating, "our conviction in the multiyear megatrend of AI remains very high." For now, the market has chosen to believe the factory manager.