NVIDIA Locks In Substrates, Packaging and Optics as AI Capacity Tightens

Capacity reservations across substrates, advanced packaging, semiconductor testing and optical engines show that the next AI bottleneck is increasingly the manufacturing stack around the accelerator—not the GPU alone.

Semiconductor Supply Chain Update | October 28, 2026

The AI supply-chain conversation has often started with accelerator availability. NVIDIA’s latest capacity agreements show why that view is now too narrow. A GPU cannot ship at scale without substrates, advanced packaging, test capacity, and increasingly optical engines that support the network architecture around the accelerator.

The scale of the commitments also matters. NVIDIA is not waiting for shortages to appear in final assembly. It is helping suppliers expand factories, reserving future capacity with advance payments, and diversifying suppliers across Japan, South Korea, Taiwan, the United States and Singapore.

For procurement teams, this is a useful warning: the most important capacity may already be allocated before downstream lead times visibly spike. Availability risk is moving from individual part numbers toward entire manufacturing ecosystems.

01 | Substrates Are Becoming Strategic Capacity

NVIDIA says agreements with IBIDEN, Samsung Electro-Mechanics and Unimicron will help drive a 2.5x increase in substrate capacity over the next three years. IBIDEN’s Ono plant is targeting three times current capacity, while Unimicron’s expansion begins adding capacity from 2Q27 and is expected to reach full capacity by mid-2028.

This matters because advanced package substrates are not generic PCBs. High-end AI packages require tighter specifications, larger formats and increasingly demanding materials. TrendForce continues to flag ABF substrates as a very tight category with little near-term relief.

02 | Advanced Packaging Is Scaling With the Accelerator Roadmap

NVIDIA and Amkor plan to expand advanced packaging capacity in South Korea, with interposer-wafer capacity targeted to increase by more than 300% by the end of 2028. This reflects the growing physical complexity of AI processors, where the package increasingly integrates logic, HBM and high-speed interconnects.

For buyers, advanced packaging should therefore be monitored as its own supply category. A processor can have foundry wafer availability and still face shipment constraints if interposer, assembly or package-test capacity is not ready at the same time.

03 | Prepayments Show How Early Capacity Is Being Reserved

One of the strongest signals is the use of capacity-reservation advances. NVIDIA is providing US$1.1 billion to KYEC to support new U.S. test facilities and equipment, with high-volume production targeted for April 2027. NVIDIA also expects total semiconductor test capacity to double from 2026 to 2027 across the U.S., Taiwan and Singapore.

Advance payments change the market dynamic. They reduce the amount of future capacity that remains freely available to other buyers and give suppliers greater confidence to commit CapEx years before final end-market demand is visible.

04 | CPO Moves Optical Engines Into the Core Supply Chain

NVIDIA and SPIL are targeting a 15x increase in optical-engine manufacturing capacity by mid-2027, with another doubling by the end of 2028. The scale of that target shows how quickly optics is moving from peripheral networking hardware into the package-level AI infrastructure roadmap.

This reinforces a theme already visible across the market: as AI clusters scale, data movement is becoming as strategic as compute. Optical engines, coupling, testing and related packaging processes should be treated as long-lead infrastructure, not just networking accessories.

05 | AI Supply Is Becoming a Portfolio of Locked Capacity

The emerging model is not simply “buy more chips.” Large AI customers are increasingly reserving capacity across multiple upstream stages at the same time. That can improve execution for the customer making the commitment, but it can also reduce flexibility for the rest of the market.

For OEMs, EMS providers and procurement teams outside the hyperscaler tier, the implication is important: future shortages may appear first as allocation, shortened quote validity, supplier prioritization and longer qualification windows—before a component is officially described as unavailable.