TSMC posts record monthly sales as Korea’s HBM buildout tightens AI hardware timelines

Taiwan Semiconductor Manufacturing Co reported record June revenue in a Taiwan Stock Exchange filing this week, underscoring unrelenting demand for AI server capacity. In Korea, investor attention stayed on Samsung Electronics and SK Hynix as both pushed next generation HBM output and capacity plans that will set delivery timelines for Nvidia and AMD through 2026.

Why it matters now, AI GPU lead times still hinge on two Asia controlled choke points, HBM stacks and advanced packaging such as TSMC’s CoWoS. Cloud and enterprise buyers planning 2027 infrastructure are watching how quickly Korea can add high yield HBM supply and how fast TSMC can clear its packaging backlog.

TSMC said June sales rose both month over month and year over year, lifting second quarter revenue to a company high. Management has previously guided that CoWoS capacity would continue to expand this year after a sharp increase in 2024, with CFO Wendell Huang telling investors the company is adding tools and sites to support more AI accelerators.

SK Hynix remains the primary supplier of HBM3E for Nvidia’s current platforms, according to company disclosures and customer tear downs tracked by Taiwanese and Korean trade press. CEO Kwak Noh Jung has repeatedly highlighted HBM mix and yield as margin drivers, and local reports this week pointed to continued pull from US customers for late 2026 deliveries.

Samsung Electronics signaled progress on HBM3E ramps and next node development while it seeks broader validation with US GPU makers. The company has paired that with investments in advanced packaging lines in Korea to complement outsourced capacity, aligning with government incentives under the K Chips framework that aim to accelerate domestic buildout.

Japan remains an important secondary node for memory and packaging. Micron, which began HBM assembly work in Hiroshima with support from the Ministry of Economy, Trade and Industry, has been layering capacity to support AI programs at US cloud providers. Japan’s subsidies for advanced logic and memory, alongside data center power initiatives, continue to pull supply chain activity to Kyushu and Hokkaido.

For global technology teams, the takeaway is that AI hardware timing is still being set in Hsinchu, Icheon, and Suwon. Procurement plans that depend on faster GPU availability will track TSMC’s packaging throughput and Korea’s HBM yields more than any single GPU product announcement.