TSMC reported NT$706.56 billion in second-quarter net income, 77.4% more than a year earlier, and raised its 2026 capital budget to a range of $60 billion to $64 billion. Revenue reached NT$1.27 trillion. The quarter records demand and profit now; the spending range describes capacity the company still has to build, equip, qualify and run. [1]
The paper's July 10 packaging account warned that a larger 2027 CoWoS target did not establish present output, factory location, customer allocation or final equipment orders. July 16 supplies a much stronger current-demand receipt and a larger capital commitment. It still does not clear today's packaging queue.
One verified X post supplies one of the edition's three observed social frames. It said TSMC's results showed that AI demand had not disappeared because cloud providers kept ordering advanced chips, infrastructure spending remained strong and demand for high-end semiconductors exceeded expectations. That is a close reading of the visible post, not an X consensus and not evidence for any financial figure in this article. The figures come from CNBC and IBTimes. [1] [2]
A profitable quarter is the first stage
Net income increased 23.4% from the previous quarter as well as 77.4% from a year earlier. Revenue rose 36% year over year. Both revenue and profit exceeded analyst estimates reported by CNBC, and the net-income figure marked TSMC's fifth consecutive quarterly record. [1]
The product mix explains why the bullish demand frame has force. High-performance computing, which includes AI accelerators and data-center processors, accounted for 66% of revenue. Process nodes at 7 nanometers or smaller supplied 77% of wafer revenue; 5-nanometer technology contributed 33% and 3-nanometer technology 30%. [1] [2]
Those percentages show what TSMC sold in the quarter. They do not identify each customer's order, distinguish a binding multi-year commitment from near-term demand, or reveal how much purchased capacity became installed systems serving paying users. High-performance computing is broader than one model company, and an AI label does not turn every wafer into measured end-user productivity.
The quarter also left customer allocation undisclosed, so strong aggregate demand cannot show which buyers hold capacity or when their packaged chips will ship under the reported multi-year investment timetable.
The capital budget is similarly real and incomplete. Chief Financial Officer Wendell Huang put this year's range at $60 billion to $64 billion as the company invests to support customer growth. A budget authorizes spending. It is not a purchase order for every tool, a completed clean room, an achieved yield or a shipped advanced package. [1]
TSMC also announced an additional $100 billion of planned Arizona investment, bringing its stated total in the state to $265 billion. Chairman C.C. Wei said the plan covers several or more logic wafer fabs for 2-nanometer mass production and advanced packaging facilities serving U.S. customers. He also described the manufacturing sequence as one that takes about five years rather than an ordinary purchase that can be made at a convenience store. [1] [2]
That horizon is the important correction to a single-quarter reading. Construction must precede tool installation. Tool installation must precede qualification. Qualification must precede stable yields. Packaging capacity must meet the wafer output. Customer allocation must then turn usable capacity into shipments. A dollar committed at the beginning of that sequence cannot be counted again as output at the end.
Demand and capacity run on different clocks
TSMC forecast third-quarter revenue of $44.6 billion to $45.8 billion and said AI-related demand remained extremely robust. Those are company expectations reported by CNBC, not independent customer-order disclosures. The locked sources do not name binding volumes by Nvidia, Apple, Broadcom or any other customer. [1]
IBTimes reported that overseas manufacturing expansion in the United States and Japan was expected to reduce margins by 2 to 4 percentage points in the near term. That is the cost side of geographic expansion: duplicated supply chains, early utilization and new-fab expenses arrive before mature output. It is also a forecast rather than a realized final margin for the facilities now planned. [2]
The sequence matters because the same facts can support two superficially opposite market stories. Record profit and a larger budget suggest confidence in demand. Heavy spending and years of construction suggest that today's scarcity cannot be solved immediately. Both are true. Neither proves that the AI cycle will continue unchanged through the life of every fab.
The verified post captures the first truth. Cloud providers and technology companies are spending enough to produce a record quarter at the world's largest contract chipmaker. It leaves out the second: advanced manufacturing expands through long, capital-intensive stages, and present orders can coexist with present bottlenecks.
The next useful receipts will be prosaic. How much of the $60-$64 billion goes to wafer fabrication, advanced packaging and other capacity? Which equipment orders are final? When do Arizona facilities move from site work to installed tools, qualification and mass production? What yields do they achieve? Which customer commitments endure, and how much overseas cost remains after utilization rises?
Until those answers arrive, the quarter should not be asked to prove more than it can. TSMC earned record profit from strong current business and raised a very large capital budget. That validates demand in the period reported. It does not make future fabs present, turn a budget into CoWoS output or guarantee that every customer queue clears on schedule.
The $64 billion in the headline is the top of a sourced $60-$64 billion range. It is a commitment ceiling for 2026, not a completed investment total. The distinction is small in typography and enormous in manufacturing. [1] [2]
-- DAVID CHEN, Beijing