TSMC's June revenue jumped 68% — and the AI chip bottleneck just moved
TSMC posted $13.8 billion in revenue for June 2026 alone — a 67.9% jump year-on-year, per its SEC filing. That single month beat most companies' annual turnover. For the first half of 2026, the Taiwan-based foundry pulled in $74.99 billion, up 35.6% on the same period last year — and it already overshot its own Q2 guidance of $40.2 billion by nearly $9 billion.
The AI engine
The driver is straightforward: AI infrastructure spending shows no sign of slowing. TSMC's 3-nanometer (N3) production lines — the most advanced in mass production — are fully booked by Nvidia, Apple, and AMD. Every high-end GPU training the models behind your AI assistant, and every chip inside the latest iPhone, almost certainly came out of a TSMC fab in Taiwan.
CNBC reports that TSMC now holds 73% of the global foundry market, according to Counterpoint Research figures for Q1 2026. Its nearest rivals, Samsung and Intel Foundry, share less than 15% between them. CEO C.C. Wei has described AI demand as "extremely robust" and said the company cannot fill orders fast enough — a situation he expects to persist for years.
The bottleneck shifts
The story here isn't just record revenue. The supply constraint has quietly moved upstream. Wafer production capacity is no longer the choke point — advanced packaging is. Specifically, TSMC's CoWoS technology, which stacks compute dies alongside high-bandwidth memory on a single substrate, is what Nvidia needs most for its data-center GPUs. Nvidia has reserved roughly 60% of TSMC's advanced packaging capacity for 2026 alone.
To ease that pressure, TSMC broke ground on Phase II of its Chiayi Science Park packaging campus on July 12. Phase I — two plants — entered mass production in June. Phase II adds at least two more facilities, with projected annual output of NT$300 billion and over 9,000 new jobs.
TSMC held 73% of the global semiconductor foundry market in Q1 2026. Illustration: Counterpoint Research
What this means outside Taiwan
TSMC's Arizona fab expansion ($20 billion investment, covering both wafer production and packaging) is approved and under construction, as is a plant in Japan. A Germany fab is also planned, though no timeline or investment figure has been published. For now, the world's AI supply chain runs almost entirely through one island — a dependency that Washington, Brussels, and Tokyo are all working to reduce, with limited speed.
The packaging bottleneck is unlikely to clear before 2027. Until then, demand for AI hardware will keep running faster than the industry's ability to assemble it.