On October 21, 2026, at its GTC Berlin developer conference in Germany, NVIDIA announced plans to expand manufacturing capacity with six supply chain partners. The plans cover substrates, advanced packaging, test and optical engines — all steps that follow front-end wafer processing.

The six partners are Amkor Technology, IBIDEN, King Yuan Electronics (KYEC), Samsung Electro-Mechanics, Siliconware Precision Industries (SPIL) and Unimicron. CEO Jensen Huang said that "demand is extraordinary and capacity is the constraint."

Plans by process step

  1. February 2027
    Singapore test expansion enters volume production

    NVIDIA's total test capacity is expected to double from 2026 to 2027.

  2. April 2027
    KYEC's US test site enters volume production

    NVIDIA backs the new site and equipment with a $1.1 billion capacity reservation prepayment.

  3. Q2 2027
    Unimicron (Taiwan) adds substrate capacity

    Expansion of the KF plant, reaching full operation in mid-2028.

  4. Mid-2027
    SPIL's CPO optical-engine capacity reaches 15x

    Expansion in Taiwan, with a further doubling targeted by end-2028.

  5. Q1 2028
    Additional capacity at IBIDEN's Ono Plant begins ramping

    NVIDIA's multibillion-dollar prepayment triples the plant's current capacity.

  6. End-2028
    Amkor (South Korea) interposer capacity up more than 300%

    Alongside development of a new manufacturing site in South Korea.

For substrates, capacity for NVIDIA products will rise 2.5x over the next three years. IBIDEN is expanding at its Ono Plant in Japan, Samsung Electro-Mechanics at its existing sites in South Korea and new plants in Sejong and Busan, and Unimicron at its KF plant in Taiwan. Samsung Electro-Mechanics is a new addition to NVIDIA's substrate supply chain, and by 2028 production for NVIDIA is expected to account for a substantial share of the company's annual package substrate capacity.

In advanced packaging, NVIDIA and Amkor aim to increase interposer wafer capacity in South Korea by more than 300% by the end of 2028. The interposer is the base layer in 2.5D packaging that places the GPU and HBM side by side and connects them, and NVIDIA says the capacity will support higher production of its current and next-generation platforms. Amkor is also expanding its advanced packaging site in Arizona (see Amkor's Arizona expansion).

The customer funds supplier capacity

What stands out in this announcement is that NVIDIA is putting money directly into its suppliers' capital spending. It is providing a multibillion-dollar capacity reservation prepayment to IBIDEN and $1.1 billion to KYEC, and has signed a capacity reservation agreement with Unimicron.

The other feature is geographic spread. Substrates are split across Japan, South Korea and Taiwan; packaging is in South Korea; and test is in the United States, Taiwan and Singapore. NVIDIA describes the aim of the plan as increasing supply chain resilience and redundancy.

What it means for procurement and business planning teams

  • Substrate and back-end capacity gets locked in early: the expansions at IBIDEN, Samsung Electro-Mechanics and Unimicron are premised on contracts with NVIDIA. Companies buying non-AI package substrates from the same makers should confirm their allocations and lead times early (for ABF substrate expansions, see AT&S and IBIDEN capacity)
  • More sources for 2.5D packaging: Amkor's interposer build-out in South Korea supplements 2.5D packaging capacity that has centered on TSMC's CoWoS (see the CoWoS/2.5D packaging race). Outsourcing the interposer step to an OSAT becomes a more realistic option
  • Optical engines as the new bottleneck: CPO optical engines carry the large 15x expansion target. For network equipment and optical component makers, securing assembly capacity becomes an issue
  • What has not been disclosed: total investment by each company, the location and start date of Amkor's new Korean site, and the terms with Samsung Electro-Mechanics

Referenced Fact Cards