November 19, 2026

AIincider

AI News. No Noise. Just Signal.

Nvidia Expands Taiwan Supply Chain: Kinsus, SPIL Ramp AI Capacity

3 min read
Nvidia is widening deals with TSMC, Foxconn, SPIL and Kinsus to lift substrate, packaging and test capacity for AI servers through 2029. Read the breakdown.

Nvidia said on Tuesday it is widening its partnerships with Taiwanese suppliers including TSMC, Foxconn, Kinsus and SPIL to lift production of AI infrastructure, Taiwan News reports, citing CNA and the company’s announcement. The Nvidia Taiwan supply chain push covers substrates, advanced packaging, testing and contract manufacturing, and it comes with concrete capacity targets that run out to 2029.

Why the Nvidia Taiwan supply chain is the bottleneck

Taiwan builds most of the world’s leading edge AI accelerators. TSMC fabricates Nvidia’s GPUs, packagers such as SPIL mount them onto interposers and substrates, and Foxconn turns them into servers and racks. The tightest constraints have moved downstream from the wafer to advanced packaging, ABF substrates and final test, which is exactly where this week’s agreements land.

What Nvidia and its partners committed to

Substrate maker Kinsus plans to raise advanced ABF substrate capacity at its Taoyuan plant by more than 3.5 times under a new agreement with Nvidia, with production expected to start in the second half of 2027 and the added capacity targeted for the third quarter of 2028. SPIL, the packaging and test unit of ASE Technology, signed a multiyear expansion adding lines for bumping, interposer wafers, assembly and testing, with that capacity expected by mid 2029. SPIL also plans to grow optical engine capacity to 15 times current levels by mid 2027 and double it again by the end of 2028.

Nvidia’s own targets are to reach 2.5 times today’s total substrate capacity within three years and to double semiconductor testing capacity between 2026 and 2027. TSMC senior vice president Chang Hsiao-chiang said the cooperation would help prepare capacity for Nvidia’s platforms, and Foxconn chairman Young Liu said his production network would support mass production of next generation AI infrastructure.

Why it matters

The announcement arrives in the same week that Taiwan’s Commercial Times reported TSMC is setting per customer credit limits and asking some chip designers to prepay for reserved capacity, a report that 24/7 Wall St. notes TSMC has not confirmed. Read together, the two stories describe a supply chain in which Nvidia is locking in capacity years ahead while smaller custom silicon designers face tighter financing terms for the same fabs and packaging lines.

For Taiwan the deals are a vote of confidence after a volatile week in which the Taiex fell sharply on AI valuation worries and Jensen Huang’s exit from Nvidia prompted local debate about the island’s AI future. The new capacity is still two to three years out, so watch for Kinsus and SPIL spending updates, TSMC’s packaging expansion figures, and whether Nvidia’s next platform ramp lands on this schedule.

Continue Reading…

Leave a Reply