TSMC's Record Quarter Shows AI Chip Demand Is Still Outrunning Supply

TSMC, the contract manufacturer behind most of the world's advanced AI chips, reported record second-quarter results on July 16 and raised both its 2026 revenue forecast and its capital spending plan. The company now expects US dollar revenue to grow slightly more than 40% this year, up from a forecast of more than 30%, according to Hargreaves Lansdown. For enterprise leaders, the results answer one question and sharpen another. Demand for AI computing is not cooling. The open question is whether the supply being built will arrive at prices and volumes that make broad enterprise deployment affordable.

What did TSMC report for the second quarter?

Revenue reached about US$40.2 billion for the quarter ended June 30, up 34% from a year earlier in dollar terms, and net income rose 77% to NT$706.6 billion, both above analyst expectations, according to Hargreaves Lansdown. Gross margin was 67.7%, the highest in more than two decades. In local currency, revenue was NT$1.27 trillion, up 36%, and advanced processes at 7 nanometers and below made up 77% of wafer revenue, Financial News reported, crediting strong AI chip orders from Nvidia, Apple and Broadcom.

TSMC also raised its 2026 capital expenditure budget to $60 billion to $64 billion and announced another $100 billion for manufacturing in Arizona, bringing its planned US investment to $265 billion, according to Hargreaves Lansdown. CFO Wendell Huang attributed the quarter to strong demand for leading-edge process technologies and pointed to a steep ramp of 2-nanometer production in the third quarter.

Is AI chip demand still outrunning supply?

The evidence this week says yes. A day earlier, lithography supplier ASML raised its own full-year outlook and said customers were accelerating capacity plans, as we reported in our coverage of ASML's results. When the company that makes the machines and the company that runs them both lift spending at once, it suggests their largest customers have committed to multi-year buildouts, not short bursts of buying.

TSMC's margin also tells a story. A 67.7% gross margin is not what a supplier earns in a buyer's market. It reflects pricing power on the most advanced nodes, where there are few alternatives. Chairman C.C. Wei told analysts that "the market continues to evolve at a rapid pace," according to DigiTimes, which also reported that CPU demand is rebounding in AI data centers, a sign that AI infrastructure needs more than accelerators.

What are the reasons for caution?

There are fair counterarguments to reading the quarter as an all-clear.

Concentration. A small number of chip designers and cloud buyers account for much of the AI demand. If one or two of them slow orders, the effect on the supply chain would be quick. The reported investor reaction was muted, with Hargreaves Lansdown describing the stock as roughly flat in Taipei despite the beat, a sign that strong results were already expected.

Capital intensity. Raising capex to as much as $64 billion is a bet that demand persists for years. Fabs take time to build, and the chips they produce will be priced to earn back that spending. Enterprises should not assume that more supply automatically means much cheaper compute.

Geography. Most of TSMC's leading-edge output is still produced in Taiwan. The Arizona expansion spreads that risk over time, but it does not remove it in the near term.

What does TSMC's outlook mean for enterprise AI budgets?

For most enterprises, TSMC is two or three steps removed. They buy cloud services, which buy servers, which contain chips that TSMC makes. But conditions at the foundry flow downstream in three ways.

Pricing will stay firm at the top end. As long as leading-edge capacity is tight and margins are high, the newest accelerators will carry premium prices, and cloud providers will pass those costs into GPU instance pricing and AI service fees. Enterprises that assume steep annual declines in AI unit costs may be disappointed.

Efficiency is the lever enterprises control. The answer is not to wait for cheaper chips but to use fewer of them. That means matching model size to the task, caching and reusing results, and turning off idle capacity. Mistral AI's Aaron Delp made the case for picking "the right model at the right time in the right place," in a AI Connect webinar on right-sized enterprise models. Bank of America's Jacob Kosoff, managing director and data science executive, put it more plainly at COLLIDE 2024: "If you build me a spaceship, I'm not going to be very happy," he said, when a simpler tool would do the job, as covered in our report on AI ROI across the project life cycle.

Capacity planning becomes procurement strategy. With supply contracted years ahead, enterprises that need guaranteed GPU access for production workloads should treat cloud commitments and private infrastructure decisions as strategic, not routine. Teams weighing on-premises AI factories against cloud should factor in lead times as well as price.

The bottom line for enterprise leaders

TSMC's quarter confirms that the AI buildout is accelerating at its foundation. That is good news for availability over the next two years, as 2-nanometer production ramps and new capacity comes online in Taiwan and Arizona. It is less clearly good news for cost. The suppliers with pricing power are using it, and the biggest buyers are locking in output first.

The enterprises best placed are the ones that already know what each AI workload costs and what it returns. Watch the hyperscalers' upcoming capital spending updates and cloud GPU pricing. If both keep rising together, the window for treating AI compute as cheap and plentiful has not yet arrived.

Frequently asked questions

What did TSMC report for Q2 2026?

Revenue of about US$40.2 billion, up 34% in dollar terms, net income of NT$706.6 billion and a 67.7% gross margin, according to Hargreaves Lansdown.

Will TSMC's results make AI computing cheaper for enterprises?

Not quickly. High margins and tight leading-edge capacity point to firm pricing for the newest accelerators, which cloud providers pass through to customers.

How much is TSMC investing in 2026?

It raised its 2026 capital spending budget to $60 billion to $64 billion and plans an additional $100 billion in Arizona, for $265 billion in total US investment.

Filed under: Data & AI Infrastructure, AI Strategy & ROI