NEWS / NOV.2026
Nvidia would commit billions to help SPIL assemble more AI chips, Reuters reports
Nvidia would fund new facilities at SPIL to assemble AI processors with their memory, Reuters reports. Even when both components are ready, a server manufacturer may still have to wait for an assembly line to become available.
Nvidia would commit billions to assemble more AI chips in Taiwan
Nvidia would provide several billion US dollars to Siliconware Precision Industries (SPIL), a Taiwanese subsidiary of ASE Technology Holding, to expand SPIL’s capacity to assemble AI chips. The money would pay for fitting out and equipping new premises, according to an industry executive familiar with the matter who spoke on condition of anonymity in a Reuters report dated November 6, 2026.
Nvidia did not respond to Reuters’ request for comment. ASE told Reuters that it does not comment on market rumors about its customers. The exact amount and form of the financing remain unknown, as do the date of the commitment and the payment schedule.
Demand for AI chips exceeds available assembly capacity, Reuters reports. For ASE, the challenge starts with finding space for the machines. On October 29, 2026, Joseph Tung, the group’s chief financial officer, said ASE was looking for suitable factories to buy. He also mentioned possible financing arrangements with customers, without naming Nvidia or giving a figure.
On January 16, 2025, Nvidia CEO Jensen Huang and SPIL Chairman C.W. Tsai attended the opening of the Tan Ke plant in Taichung. In a press release that day, SPIL said it had designed the plant to meet Nvidia’s demand and had already begun increasing production there. Reuters did not specify which country would host the new facilities Nvidia would fund.
Why a server may have to wait for its accelerator
The new facilities would use CoWoS, an assembly process developed by TSMC. The processor and the stacked high-bandwidth memory, known as HBM, sit side by side. A layer directly beneath them connects the two so they can exchange data quickly. The entire assembly rests on a support called a substrate, as shown in TSMC’s diagram.
Even when the processor and memory are ready, an assembly line still has to put them together before the accelerator can be installed in a server. The wait can delay the date when a data center brings its machines online and, in turn, when a lab can rent computing power there.