TSMC reached the top of its second-quarter revenue guidance and raised its 2026 outlook, but a one-off investment gain boosted profit growth while the company committed more capital to 2-nanometer production, advanced packaging and an undated Arizona expansion.
TSMC's record quarter strengthened the case that AI infrastructure spending is concentrating at the world's largest contract chipmaker. It also exposed the cost of that position: more capital, a near-term margin penalty, tight packaging capacity and a U.S. expansion whose timing still depends on customers.
TSMC reported NT$1.27038 trillion in revenue for the three months ended June 30, up 36% in Taiwan dollars from a year earlier and 12% from the first quarter. In U.S. dollars, revenue rose 33.7% year over year to $40.2 billion. That was the top—not a beat—of the company's $39 billion-$40.2 billion guidance.
Net income attributable to shareholders rose 77.4% year over year and 23.4% sequentially to NT$706.56 billion. Diluted earnings were NT$27.25 per share, or $4.31 per American depositary receipt. Independent reporting on the results described both quarterly revenue in Taiwan dollars and net profit as records.
Operations were highly profitable. Gross margin was 67.7%, 0.2 percentage point above the top of the range in TSMC's earnings presentation. Operating margin reached 60.3%, compared with 49.6% a year earlier.
But the bottom-line comparison includes a large non-operating boost. TSMC recorded NT$95.83 billion of non-operating gains, up from NT$29.61 billion a year earlier. Its earnings-call transcript attributes NT$63 billion to disposal and mark-to-market gains on Vanguard shares, adding NT$2.24 to earnings per share. The revenue and operating-margin gains remain substantial, but 77.4% is not a clean measure of growth in the chipmaking business.
High-performance computing grew 20% from the first quarter and supplied 66% of revenue. Smartphones, once TSMC's largest platform, fell 4% sequentially to 22%. The high-performance-computing category includes more than AI, but leading AI accelerators and data-center processors sit inside it.
The technology mix moved in the same direction. Five-nanometer products accounted for 33% of wafer revenue, 3-nanometer for 30%, 7-nanometer for 11% and the ramping 2-nanometer process for 3%. Processes at 7 nanometers or below together generated 77% of wafer revenue.
Chairman and Chief Executive C.C. Wei said cloud providers were giving TSMC very strong demand signals. He also said agentic AI was increasing the role of CPUs alongside accelerators, benefiting TSMC because leading x86, Arm and RISC-V suppliers are almost all its customers. That is a company forecast, not a measured contribution from agentic AI to current revenue.
The rest of the market is weaker. Wei said consumer and price-sensitive segments face rising component prices and macroeconomic uncertainty. At mature nodes, AI-linked power-management chips and sensors are in shortage, while demand in other commodity segments is not strong. An outside analyst similarly said rising memory prices and tight component supply were hurting TSMC's non-AI business in an account of the results.
That split narrows the claim that TSMC is benefiting from a semiconductor-wide boom. Its strongest demand is concentrated in leading-edge compute and the mature-node components attached to AI systems.
TSMC lifted its 2026 capital budget to $60 billion-$64 billion from $52 billion-$56 billion. It plans to direct about 70%-80% to advanced processes, about 10% to specialty technologies and 10%-20% to advanced packaging, testing, maskmaking and other spending. Wei said higher demand was the main reason for the increase; inflation in tool prices was the second.
The higher budget is not proof that every customer forecast will convert into installed systems. Wei said TSMC combines bottom-up and top-down forecasts because each customer's aggressive projection may be reasonable alone but cannot simply be added to all the others. The company checks data-center locations, construction progress and rack deployment to reduce the risk that its chips accumulate in inventory.
Packaging is already limiting how many finished systems customers can ship. Wei said the gap between packaging demand and capacity was “very big.” He welcomed alternatives such as Intel's EMIB-T because they could package more TSMC-made wafers. The distinction matters: competing packaging can take back-end work from TSMC while relieving a bottleneck in its much larger front-end wafer business.
The foundry business also has alternatives. When an analyst raised Samsung, Intel and reports that customers were engaging with competing manufacturers, Wei argued that technology qualification, production ramping and customer trust make switching a roughly five-year process. That response explains TSMC's customer stickiness, but it does not establish that rival capacity is irrelevant.
Pricing shows the same trade-off. An analyst said TSMC had more pricing power than it was using. Wei said the company would not impose abrupt increases that could price customers out of their own markets; its stated goal is to earn enough to sustain long-term expansion. On Wei's account, TSMC is accepting less near-term pricing leverage to preserve customer economics.
The new capacity will initially pressure margins. TSMC expects the steep 2-nanometer ramp to reduce gross margin by about 3-4 percentage points in the second half. Overseas-fab expansion is expected to dilute gross margin by 2-3 points in its early stages and 3-4 points later. Third-quarter gross margin guidance of 65%-67% is below the second quarter's 67.7%.
Investors have set a demanding comparison base. TSMC's New York-listed shares closed 2.3% lower after the report, according to market data. A fund manager said in an access-limited report that expectations were “exceptionally high,” while still crediting the company's execution and technology position.
TSMC said it would invest another $100 billion in Arizona, raising its planned U.S. investment to $265 billion. The company said the money would support several more logic fabs for 2-nanometer and more advanced processes, along with advanced-packaging facilities for leading U.S. customers.
Asked for a facility count, Wei said “probably additional four or more fabs” across front- and back-end production. Asked twice for a timetable, he declined to give one. The pace will depend on market conditions and customer demand, even as TSMC tries to accelerate construction. Independent reporting on the announcement also noted both the margin cost of overseas expansion and the absence of a firm schedule. A separate account of Wei's answers described the count as four more fabs, combining front- and back-end facilities.
That makes $100 billion a destination for capital rather than a dated construction program. Some coverage characterized the new money as funding four additional advanced-chip plants, but TSMC's own wording was less fixed: “four or more,” including packaging, with no firm schedule.
The industrial-policy financing predates this new commitment. For the earlier $65 billion, three-fab plan, the U.S. Department of Commerce awarded TSMC Arizona up to $6.6 billion in direct funding and offered up to $5 billion in loans. The 2024 award ties direct disbursements to construction, production and commercial milestones. Public support therefore lowers financing pressure only as the earlier project meets government conditions.
TSMC is not describing Arizona as a replacement for Taiwan. Wei said the company is building 13 leading-edge and advanced-packaging fabs in Taiwan over the next several years. It is also adding one 3-nanometer fab in each of Taiwan, Arizona and Japan, while converting some 5-nanometer tools in Taiwan to support 3-nanometer output.
The most detailed retained infrastructure record covers the original three-fab Arizona plan, not the newly announced facilities. That distinction prevents its estimates from being presented as a forecast for the full $265 billion commitment.
A 2024 draft environmental assessment estimated that the first three phases, operating at full volume, would require 17.29 million gallons of city water and 8.54 gigawatt-hours of electricity per day. The existing development agreement covered up to 11.4 million gallons of water per day; the assessment said TSMC Arizona would need a revised or new agreement with Phoenix for up to 17.3 million.
The same assessment said an industrial water-reclamation plant had to begin operating by June 2028 under the wastewater permit and was designed to recycle at least 95% of wastewater. Its full-volume resource table estimated average discharge of 13.83 million gallons per day across the three phases, while the project analysis said pretreated wastewater would go to the city's treatment system. TSMC Arizona had applied for and mostly obtained more than 600 phased and unphased permits for the original facility.
Those figures do not prove the new Arizona expansion lacks water, power or permits. They show that even the earlier plan depended on negotiated utility capacity, a treatment project and hundreds of approvals. The additional facilities will need their own updated engineering and regulatory record before the full infrastructure burden is knowable.
TSMC expects third-quarter revenue of $44.6 billion-$45.8 billion. At the midpoint, that would be about 12% above the second quarter and 37% above the prior-year period. It also raised its 2026 forecast to revenue growth slightly above 40% in U.S. dollar terms, from more than 30% previously.
Wei said demand could remain very strong through 2029 or 2030, but he could not rule out a dip in between. The next evidence should test the assumptions behind the spending rather than merely confirm another revenue record:
Until those measures are visible, the quarter establishes strong current demand and pricing economics, not the eventual return on TSMC's larger capacity plan.
OpenAI has put conversations and Projects back in its redesigned ChatGPT desktop app and enabled cloud Work threads to move across devices, correcting the launch's biggest usability failures without merging local Work or Codex histories.
Meta reportedly plans to put departing AWS compute executive Dave Brown to work on its data-center buildout, adding hyperscale operating experience while leaving any customer-facing cloud business conditional and undefined.
China has paired a five-year AI training offer for developing countries with cooperation centers, a weather-warning rollout and a new 29-country organization. The package gives Beijing a platform for influence, but no budget, selection rules or delivery timetable has been published.
RoboTTT extends a robot policy’s visuomotor context to 8,000 timesteps and raises its average rubric score from 42% to 79%, but the evidence comes from three author-run assembly tasks and the longest task was completed in only two of ten trials.
Zhipu reportedly reached $1 billion in annual recurring revenue in July, roughly four times a March estimate, but the unconfirmed run rate is not annual sales and still sits far ahead of recognized cloud revenue while margins remain thin.
Nebius has arranged its first senior secured facility against an operating GPU deployment and one customer’s cash flows. The deal adds project-level debt to its expansion toolkit, but an unnamed customer and a larger rival financing limit what it proves about the rest of Nebius’s backlog.
Cursor says automatically running a repository-root git.exe on Windows does not meet its criteria for patching, while the researcher calls it an untrusted-search-path defect and separate research shows the same weakness across several competing AI coding tools.
Huawei publicly displayed a 16-cabinet Atlas 950 configuration rated at 1 EFLOPS in FP8, providing tangible evidence of its system-scale AI strategy while leaving price, power use and sustained workload performance undisclosed ahead of the full system's planned fourth-quarter release.
CIA Director John Ratcliffe said US intelligence is consistent with an estimate that Russian recruits last 20 to 30 minutes on Ukraine’s battlefield, but the public trail leads to an unsourced claim about assault troops and does not establish a representative average.
UK testing places leading open-weight models four to seven months behind selected closed-model cyber results, yet longer attack chains, U.S. benchmarks and mixed cost comparisons show why that interval is a warning signal rather than a universal capability clock.
Moonshot AI's Kimi K3 added to a global technology selloff with near-frontier performance, but unreleased weights, mixed cost comparisons and a recommended 64-accelerator deployment leave its effect on chip demand unresolved.
Moonshot AI has made Kimi K3 available through its apps and API, pairing a 2.8-trillion-parameter architecture with early frontier-level results, but the model's open-weight claim cannot be tested until its weights and technical report arrive.
Two binding EU decisions require Google to give rival AI services comparable access to 11 Android features and offer eligible search competitors a restricted, anonymized dataset, but phased deadlines, certification, pricing and privacy safeguards leave the competitive effect unproven.
Databricks has signed a term sheet for a Coatue-led financing at a $188 billion valuation, while unidentified sources put the round at $3 billion. The proposed capital would deepen its push into AI governance, data agents and operational databases, but the transaction remains open and the company supplied no new operating figures.
Netflix says generative AI workflows were used on roughly 300 titles in 2026, but its only quantified example tied to that disclosure covers 17 minutes and does not establish how much finished material, spending or labor changed across the slate.
Gemini 3.5 Pro missed its expected June rollout. An anonymous-source account says a late-June data change fell short of Google's coding goals, but Google has confirmed only partner testing—not the reported cause, a new date, or public results and pricing.
Google is bringing Gemini Omni editing and a reusable face-and-voice avatar to Vids, but the sharper distinction is account-level identity across Vids and Gemini rather than a new category of AI video; Vids already offered Veo generation and customizable avatars, while specialist rivals already sell digital presenters.
Intel plans to adopt Gemini Enterprise across engineering, supply-chain and corporate work while extending chip-simulation capacity into Google Cloud, but the companies supplied no rollout schedule, contract value or performance baseline.
The Navy has approved an immediate department-wide framework for turning data into operational effects, but the public rollout leaves budgets, owners, deadlines, performance measures and assurance rules to a promised implementation roadmap.
Runta has raised a $20 million seed round led by Andreessen Horowitz for an execution layer that governs agents at the operating-system and network levels, entering a market where cloud runtimes and sandbox providers already offer overlapping controls.