NVIDIA announced plans to expand manufacturing capacity for AI infrastructure. In partnership with 6 companies, including Ibiden, it will increase package-substrate production 2.5-fold over the next 3 years.
NVIDIA announced a joint plan with 6 companies, including Ibiden, to expand production capacity for AI infrastructure. The plan centers on package-substrate capacity, while production of testing and optical communications components will also be increased to strengthen overall supply capability. The policy is to respond to rapidly growing demand by expanding 4 processes simultaneously: substrates, advanced packaging, testing and optical communications.
Specific capacity-expansion plans were presented for the 4 processes. Ibiden, Samsung Electro-Mechanics and Unimicron will handle substrates; Amkor will expand interposer capacity; KYEC will handle testing, including expansion in Singapore; and SPIL will handle optical engines. The start dates and target multiples vary by process and region.
Japan’s Ibiden plans to begin additional capacity at its Ohno plant in Q1 2028, targeting capacity 3 times current levels. South Korea’s Samsung Electro-Mechanics will begin supplying immediately, while Taiwan’s Unimicron is expected to start additional supplies from Q2 2027 and reach its target capacity in mid-2028. For semiconductor testing, NVIDIA will make advance payments to KYEC and aims to begin mass production in April 2027, including at a new U.S. facility. SPIL aims to increase optical-engine production 15-fold by mid-2027 and by an additional 2-fold by the end of 2028.
According to NVIDIA’s 10-Q, the new investment in the 6 companies is not directly synonymous with the current initiative, and the total value of contracts as of 2026 has reached $279 billion. These are multiyear supply-guarantee contracts, so the total investment amount cannot be calculated by simply adding them together. The advanced-packaging contract with Amkor is separate, and capital investment in the United States and South Korea is expected to proceed in stages. Actual production ramp-up and achievement of the targets are expected to occur at different times for each process.