Nvidia Capitalizes Advanced Packaging and Optics Supply Chains to Lock Up AI Compute Capacity
Nvidia is executing a massive supply chain strategy, securing multi-billion-dollar capacity reservations through 2028 for substrates, co-packaged optics, and OSAT services to sustain its AI accelerator roadmap.
Nvidia is aggressively restructuring its procurement strategy, shifting from standard purchase agreements to direct capital injections for physical capacity buildouts. Following announcements at GTC Berlin on October 21, 2026, the company confirmed multi-billion-dollar commitments spanning substrates, advanced packaging, semiconductor testing, and optical engines. These agreements, which extend through 2028, are designed to mitigate severe bottlenecks in the post-wafer manufacturing ecosystem. As CEO Jensen Huang noted, "We are in the midst of the largest infrastructure buildout in human history. Demand is extraordinary. Capacity is the constraint."
The substrate layer has emerged as a primary bottleneck, particularly as Nvidia scales its Blackwell, Rubin, and Rubin Ultra architectures. These designs require larger substrates to accommodate increasing numbers of GPU dies and HBM3E stacks. According to Tom's Hardware, Nvidia is targeting a 2.5x increase in dedicated substrate capacity over the next three years. This expansion is supported by capacity reservation deals with Ibiden, Unimicron, and Samsung Electro-Mechanics. Notably, Nvidia provided a multibillion-dollar advance to support Ibiden’s Ono Plant in Japan, while its agreement with Unimicron is expected to begin contributing capacity by Q2 2027.
To address power-density walls inherent in traditional copper trace networks, Nvidia is accelerating the adoption of co-packaged optics. By moving optical engines directly into the accelerator package, the company aims to maintain signal integrity at speeds exceeding 224 Gbps per lane. Nvidia has secured a partnership with SPIL to expand optical engine manufacturing capacity by 15x by mid-2027, with a further doubling planned by the end of 2028. This move signals a strategic shift toward integrating silicon photonics as a standard component in its high-performance accelerator packages, effectively bypassing the limitations of traditional board-level interconnects.
Back-end manufacturing capacity is receiving similar capital prioritization. Nvidia has committed $1.3 billion to Amkor, part of a total $3 billion investment across the OSAT’s Arizona and South Korean facilities, to increase interposer wafer capacity by over 300% by the end of 2028. Simultaneously, the company is addressing post-assembly bottlenecks through a $1.1 billion deal with KYEC. This investment targets a doubling of semiconductor testing capacity by 2027, with high-volume production slated for KYEC’s new U.S. facility and expanded operations in Singapore.
This strategy represents a significant departure from traditional fabless models, where design houses typically offload capital risk to suppliers. By July 2026, Nvidia had accumulated $279 billion in outstanding supply and manufacturing commitments, reflecting a comprehensive effort to control the entire production stack—from fabrication and HBM memory to final assembly and testing. While Nvidia dominates the AI accelerator market, competitors are responding; AMD announced over $10 billion in investments across the Taiwanese semiconductor ecosystem on May 21, 2026, specifically targeting advanced packaging and manufacturing partnerships to increase production in 2027.