Nvidia backs Taiwan suppliers with billions for AI capacity

Nvidia is committing billions of dollars to help Taiwan suppliers expand the manufacturing capacity its future AI systems will need. New agreements with substrate producer Kinsus and packaging specialist Siliconware Precision Industries, or SPIL, show how the chip designer is using funding and multiyear supply commitments to secure production well beyond its current product cycle.
Nvidia’s November 17 announcement includes plans to expand Kinsus’s Nvidia-dedicated substrate capacity to more than 3.5 times its current level. SPIL, a subsidiary of ASE Technology Holding, will receive multibillion-dollar Nvidia funding for advanced packaging expansion in Taiwan.
Kinsus said Nvidia would provide more than US$1 billion in cash for equipment at its Youshi campus in Taoyuan. Production is expected to begin in the second half of 2027, with additional capacity targeted for the third quarter of 2028. Kinsus also said discussions were continuing over investment at other sites to support longer-term supply.
Package substrates provide the fine electrical connections between a semiconductor package and the circuit board beneath it. As AI packages become larger and incorporate more components, their substrates require more layers and increasingly complex structures. Kinsus says those changes are increasing demand for manufacturing capacity as well as raising technical requirements.
For the supplier, Nvidia’s funding helps finance equipment purchases against a defined customer requirement. For Nvidia, it secures a dedicated source of components that can otherwise constrain shipments.
SPIL’s expansion addresses another production constraint. Advanced packaging connects processors and high-bandwidth memory within a single package, allowing data to move quickly between them. SPIL said its new multiyear agreement targets a tripling of the relevant advanced packaging capacity by mid-2029, with investment in production lines, factory infrastructure, and automation.
The planned operations cover bumping, which creates tiny electrical contacts on chips; interposer-wafer fabrication, which provides dense connections between components; assembly; and testing. Handling all these steps lets SPIL produce more of each package for Nvidia.
This agreement is separate from the optical-engine expansion announced in October. That earlier project supports co-packaged optics, which places optical components close to chips for high-speed communication. Nvidia’s October commitments also included a US$1.1 billion capacity-reservation advance to King Yuan Electronics for new US testing facilities and equipment. Across its partners, Nvidia targeted a doubling of semiconductor testing capacity between 2026 and 2027.
Nvidia's announcement also pointed to the Taiwan companies that make its chips and servers. It said Vera Rubin NVL72 systems and Vera CPUs for agentic AI began volume shipments in the third quarter, using TSMC's 5nm and 3nm processes and CoWoS advanced packaging. Foxconn stated that its production network across Asia, North America, and Europe would support the ramp-up of Nvidia's next-generation GPU and networking platforms.
Unlike TSMC and Foxconn, whose existing roles the announcement confirms, Kinsus and SPIL are adding capacity that does not yet exist. They must order machinery, build factories, and qualify processes long before that capacity earns revenue. With their schedules reaching into 2028 and 2029, Nvidia's ability to deliver AI systems for several years now depends more directly on Taiwan's suppliers.
