On November 17, 2026, NVIDIA announced multiyear agreements with Taiwan's Kinsus and SPIL, along with a revised substrate supply agreement with South Korea's Samsung Electro-Mechanics, to expand production of AI semiconductors. A package substrate is the base that electrically connects a chip's tiny contact points to the board it is mounted on. Advanced packaging, meanwhile, is a technology for assembling multiple chips and components with high-density interconnections. Under these agreements, the timing and targets for capacity increases differ between substrate manufacturing and advanced packaging.
Different Timelines for Substrates and Advanced Packaging
Kinsus plans to bring additional capacity for NVIDIA substrates online in the third quarter of 2028, and SPIL plans to expand its advanced packaging capacity for NVIDIA by mid-2029. According to NVIDIA's official Japanese-language post, Kinsus is expected to begin production at its Youth campus in Taoyuan, Taiwan, in the second half of 2027, with additional equipment coming online after that. Under the new multiyear agreement, the plan is to raise dedicated NVIDIA ABF package substrate capacity at that site to more than 3.5 times the current level. According to Ajinomoto's technical explanation of ABF, which the company developed, ABF is a film that insulates the layers between wiring layers when fine wiring is built up in multiple layers on a package substrate.
Based on NVIDIA's November 17 announcement and Samsung Electro-Mechanics' November 18 announcement, each company's expansion plans can be summarized as follows.
| Company / Site | Process | Target | Timing | Status of agreement / plan |
|---|---|---|---|---|
| Kinsus / Youth campus, Taoyuan, Taiwan | Dedicated NVIDIA ABF package substrates | Start of production | Second half of 2027 | Expected under the new multiyear agreement |
| Kinsus / Youth campus, Taoyuan, Taiwan | Dedicated NVIDIA ABF package substrates | Additional capacity comes online | Q3 2028 | Target under the same agreement |
| SPIL / Taiwan | Dedicated NVIDIA advanced packaging | Expansion of contracted dedicated capacity | By mid-2029 | Multiyear agreement with NVIDIA providing multibillion-dollar funding |
| Samsung Electro-Mechanics / Sejong and Busan, South Korea | Package substrate production infrastructure | Building out full-scale mass production | From 2029 | Revision of existing agreement agreed. Additional funding support under consideration by NVIDIA |
What can be compared from the official announcements of November 17–18, 2026, is the timing and target of each process's expansion. No data has been published that would allow a direct comparison of Kinsus's and SPIL's capacity by volume or utilization rate.
The production lines SPIL is adding in Taiwan will handle bumping, which forms the protruding electrodes used to connect chips, and the manufacturing of interposer wafers that connect chips to one another. The processes covered also include chip assembly and testing. SPIL plans to expand, by mid-2029, the advanced packaging capacity that NVIDIA has secured through the contract. The year 2029 cited by Samsung Electro-Mechanics, by contrast, refers to when it begins installing substrate production equipment and setting up full-scale mass production.
How Far Has Additional Funding Support for Samsung Electro-Mechanics Been Decided?
According to Samsung Electro-Mechanics' November 18 announcement, to expand the supply of package substrates for next-generation GPUs, the company plans to improve line utilization efficiency at its existing plants in Sejong and Busan, South Korea, while maintaining quality control standards. Building a setup that can flexibly adjust production volume in response to demand fluctuations is also among the aims of the revised agreement. The collaboration NVIDIA announced on October 21 also set out plans for equipment upgrades and construction of new plants in Sejong and Busan, in addition to supply from existing facilities in South Korea.
According to Samsung Electro-Mechanics, the two companies agreed in October that NVIDIA would provide funding to expand dedicated plants for AI semiconductor package substrates. In the November announcement, the company disclosed that NVIDIA is considering additional funding support to further expand production infrastructure in Sejong and Busan. The company plans to install equipment with that additional support and set up full-scale mass production from 2029. Accordingly, while the revision of the existing agreement and the funding agreed in October are settled, the additional support described in November remains under consideration.
The partnership between the two companies also extends to multilayer ceramic capacitors (MLCCs) used in NVIDIA's computing platforms. Samsung Electro-Mechanics explains that stable power supply is essential in computing environments where many components are densely mounted and consume large amounts of power. The collaboration covers not only package substrates for next-generation GPUs but also the electronic components that support the power circuits of the computing platforms.
What Is the Difference Between October's 2.5x Substrate Figure and November's 3.5x-Plus Figure for Kinsus?
The 2.5x figure given in NVIDIA's October 21 announcement refers to a plan to raise total production capacity of package substrates for NVIDIA products over the next three years. Agreements with Ibiden, Samsung Electro-Mechanics and Unimicron support this expansion. The 3.5x-plus figure announced on November 17, by contrast, compares the capacity NVIDIA is securing at Kinsus's Taoyuan site with that site's current dedicated capacity. In other words, October's 2.5x covers NVIDIA's substrates overall, while November's 3.5x-plus covers one specific Kinsus site, so the baselines for comparison differ.
For SPIL as well, the processes being expanded differ between the October and November announcements. The October 21 announcement covered optical engines used in co-packaged optics (CPO), components that support high-speed communication in large-scale computing facilities. Under the plan NVIDIA outlined, SPIL will raise optical engine production capacity 15-fold by mid-2027 and double it again by the end of 2028. The multiyear agreement announced on November 17, meanwhile, covers dedicated advanced packaging capacity spanning bumping and interposer wafer manufacturing through chip assembly and testing.
Regarding Amkor, the figure of just under $3 billion that NVIDIA presented on November 17 is the total of multiyear agreements already announced in July and October, and it includes prepayments that support equipment expansion. It covers advanced packaging capacity in Arizona in the United States and in South Korea. In South Korea, NVIDIA announced on October 21 a target of increasing interposer wafer production capacity by more than 300% by the end of 2028.
Vera Rubin Volume Shipments and Future Supply Capacity
According to NVIDIA's November 17 announcement, volume shipments of Vera Rubin NVL72 and the Vera CPU began in the third quarter of 2026. The company cites TSMC's 5nm and 3nm manufacturing processes, as well as its CoWoS advanced packaging technology, as supporting shipments of these products. Foxconn, for its part, says it is positioned to support expanded production of next-generation GPUs and networking products by leveraging its production network spanning Asia, North America and Europe.
For Kinsus, neither the current production capacity, the absolute volume after expansion, nor actual order volumes have been disclosed. SPIL has likewise not disclosed the specific size of its capacity increase or the exact amount of funding. For Samsung Electro-Mechanics, the timing and amount of the additional funding support decision, the substrate supply volume at full-scale mass production, and the MLCC supply volume and start date remain undisclosed.
Volume shipments of Vera Rubin began ahead of the expansion plans announced now. Once Kinsus's and SPIL's added lines come online, the substrate supply capacity NVIDIA can secure and the throughput from chip assembly through testing will grow. How much these equipment expansions contribute to shipments of completed systems will depend on each line's utilization rate and how capacity is allocated among products.
