Ibiden ships FC-BGA to TSMC, not NVIDIA: industry survey

The survey's starting point is a simplification it wants to dismantle: that a booming generative-AI market and rising Nvidia GPU shipments translate straight into Ibiden orders and sales. In the author's account, the supply chain works differently. Ibiden engages chip architects at Nvidia, AMD, Google (TPU) and AWS (Trainium) two to three years before a product launch, co-designing to suppress extreme warpage and transmission loss, and wins top-tier AVL designation. Only then does it feed substrates into TSMC's CoWoS line and bill TSMC.

The report frames the technical difficulty as the reason this arrangement holds. As package substrates grow past 100mm per side and beyond 20 to 30 layers, mismatched thermal expansion between silicon dies and organic substrates, ultra-low-impedance power delivery at 1,500A to 2,000A, and exponential area-yield loss all compound. A single microvia defect can scrap a finished GPU package worth tens of thousands of dollars.

The stakes hinge on whether Ibiden can defend that yield and share advantage, and on whether the arrangements it describes — capacity reservations, advance payments, and the reported TSMC alliance around glass core substrates — hold as the next generation of packaging arrives. That is the question for analysts and supply-chain planners watching this segment.