AI Demand Lifts Foundry Market 29% in Q2

Asia Business Daily reported that TSMC's share put it 66 percentage points ahead of Samsung in Q2. Counterpoint attributed TSMC's position to 2nm mass production, expansion of 3nm processes, shortages in 8-inch and 12-inch mature-node capacity, and tight supply of advanced packaging.

The reported mix matters because advanced AI accelerators depend on more than leading-edge logic fabrication. They also require packaging capacity to integrate high-bandwidth memory and compute dies. Across the sector, constraints at either the wafer-fabrication or packaging stage can limit delivered accelerator volume even when demand for chips remains high.

Counterpoint reported that surging AI-related orders and capacity reallocation remained key drivers of supply-demand imbalances in both advanced and mature nodes. Mature-node demand remains relevant to AI infrastructure because power-management ICs, connectivity components, controllers, and other supporting silicon typically do not require the newest process technologies.

Samsung's Q2 share was unchanged from Q1 2026, according to both reports. Counterpoint identified demand for Samsung's SF4 and SF5 processes, along with foundry wafer price increases in the first half of 2026, as drivers of revenue growth. The research firm also reported that Samsung's principal target remained yield improvement for its SF2 process.

SMIC recorded record Q2 revenue and retained third place, supported by domestic Chinese demand and increased overseas orders, according to Counterpoint as cited by Asia Business Daily. The report did not provide revenue figures in the material reviewed.

Intel did not appear in Counterpoint's top-10 pure-play foundry ranking, LaPedus reported. That ranking is specifically about pure-play foundry share and does not measure every semiconductor manufacturer's total chip revenue or internal manufacturing output.

Counterpoint forecast that foundry wafer average selling prices would continue rising in the second half of 2026 and that pure-play foundry utilization would remain relatively high. That is a forecast from the research firm, not an announced pricing policy by individual foundries.

For ML infrastructure teams, the market data is a reminder that AI system availability can be shaped by a multistage hardware supply chain, not solely by GPU architecture or cloud procurement. Wafer fabrication, packaging capacity, and total system availability can all affect large-scale training and inference deployments.