TSMC: June revenue +68% - compute-crunch has a very strong pulse

In plain terms

Taiwan Semiconductor achieved a 68% year-over-year revenue increase in June 2026. Translation: demand for advanced AI chips (Hopper, Blackwell, Rubin) remains significantly in excess of supply—and the waiting list continues to grow.

The context

TSMC physically manufactures most of the high-end silicon used for generative AI: Nvidia GPUs (Hopper, Blackwell, Rubin), Google TPU accelerators, Amazon Trainium, AMD MI300/400. Its monthly reporting cycles are one of the few near real-time indicators of the health of the compute market. The previous months' series already showed high double-digit YoY growth; moving to +68% in June 2026 is an acceleration, not a plateau.

The data

According to Tech in Asia (July 13, 2026), the figures published by TSMC:

• June 2026 revenue: +68% YoY.
• Cumulative revenue H1 2026: NT$2,404 billion (~US$74.8 billion).
• Q3 2026 guidance raised by management (details not reproduced in the source).

The analysis

Three signals to isolate. First: the "AI bubble forming" thesis loses an argument, as this is not announced revenue but wafers delivered and billed. Second: the product mix continues to shift towards 3 nm and 2 nm—these are the nodes for AI, not for smartphones. Third: Nvidia captures most of it, but TSMC also supplies TPU accelerators (Google) and Trainium (AWS)—the silent decentralization of AI silicon continues.

So what

For an investor: the gap between the "peak AI" narrative (cf. ETNews today on SK Hynix) and TSMC's reality (+68%) is the real end-of-summer debate. Two different markets are observing the same industry. For a compute-dependent executive: H2 2026 delivery times are not shortening—do not base a product plan on declining GPU pricing before 2027.