Foundry Revenue Hits Record as CoWoS-L Packaging Stays the Binding Constraint Through 2028

The world's top ten foundries posted combined revenue of nearly US$53.49 billion in the second quarter of 2026, up 11.5% sequentially and a fresh record, according to TrendForce. Growth was driven by continued supply constraints on the advanced processes used in AI and high-performance computing processors, together with rising demand for the peripheral chips that surround them, including power management ICs and power discretes. Advance procurement across consumer supply chains, spanning televisions, PCs and notebooks, also tightened capacity in some mature processes.

Packaging, not lithography, remains the narrowest point in the chain. TrendForce's latest advanced packaging research finds that GPU suppliers and hyperscale cloud providers are pushing ever larger package sizes, making integration beyond reticle limits the central development priority for 2.5D packaging. TSMC's CoWoS-L faces competition from Intel's EMIB-T, but its advantages in technology maturity and yield are expected to keep it the mainstream solution for AI chips through 2028.

The practical consequence is that nominal wafer capacity no longer tells buyers how much compute will actually reach the market. Even as leading foundries add front-end capacity, the assembled and tested module, combining logic die with memory stacks and high-speed I/O, is what sets the shipment ceiling. Buyers planning 2027 capacity should therefore treat packaging slots rather than wafer starts as the scarce item to lock in, because a qualified alternative for a large 2.5D package cannot be stood up quickly.

For procurement teams the record quarter is a warning rather than an all-clear. Revenue growth of 11.5% in a single quarter with advanced nodes already fully loaded implies that incremental 2027 demand will compete for a capacity base that is essentially fixed over a one to two year horizon. In that environment lead times and allocation, not spot prices, are the first variables to move when a new AI programme is committed.