Global AI Infrastructure Watch: How NVIDIA and Six Suppliers Are Rebuilding Manufacturing Capacity
Rising demand for accelerated computing and for the full-stack AI factory platform has moved NVIDIA's answer away from the roadmap and onto the factory floor. The company has unveiled a plan developed together with its supply chain partners to broaden global manufacturing capacity, and the commitments attached to it are stated in multiples rather than single-digit percentage gains: substrates, advanced packaging, testing and optical engines are all in scope.
Read as a whole, the disclosure is less about one new production line than about redundancy — a second and third source for capabilities that today sit with a small handful of manufacturers across Japan, South Korea, Taiwan, US and Singapore sites.
The headline numbers in one place
Six suppliers are named: Amkor Technology, IBIDEN, King Yuan Electronics (KYEC), Samsung Electro-Mechanics, Siliconware Precision Industries (SPIL) and Unimicron Technology Corporation. Four capability areas frame the buildout.
- Substrates: capacity set to grow 2.5x in three years — NVIDIA intends to lift substrate capacity for its products by 2.5x over the next three years.
- Testing: total semiconductor testing capacity is expected to double from 2026 to 2027, measured from calendar 2026 to 2027.
- Advanced packaging: a greater than 300% increase in interposer wafer capacity in South Korea is targeted by end of 2028.
- Optical engines: co-packaged optics capacity in Taiwan targets an increase of 15x by mid-2027, with aims for another doubling by the end of 2028.
Substrates: the 2.5x target and the three agreements behind it
Substrate supply is the first pressure point, because each package needs one before it can be assembled or tested. NVIDIA intends to lift substrate capacity by 2.5x over the next three years, and the agreements described below will help drive that expansion, with more in progress.
IBIDEN and the Ono Plant in Japan
In an agreement signed and announced earlier this year by IBIDEN, NVIDIA is extending a multibillion-dollar capacity reservation advance to support the supplier's Ono Plant in Japan. The plant expansion aims for capacity reaching 3x current levels, and additional capacity is expected to begin ramping in the first quarter of 2028. The 3x figure at a single site sits well above the 2.5x overall target, which indicates that part of the global substrate growth is concentrated in one ramp rather than spread evenly across the network.
Samsung Electro-Mechanics enters the substrate chain
Samsung Electro-Mechanics joins NVIDIA's substrate supply chain, contributing immediate capacity from its existing operations in South Korea — a distinction that matters for the near term, since established lines can be allocated sooner than greenfield plants. The collaboration also backs facility upgrades and new factories in Sejong and Busan, South Korea. By 2028, production allocated to NVIDIA is projected to represent a significant portion of Samsung Electro-Mechanics' total annual package substrate production capacity.
Unimicron and the KF expansion in Taiwan
NVIDIA is backing Unimicron in its KF substrate manufacturing expansion in Taiwan through a capacity reservation agreement. That expansion is expected to begin adding capacity in the second quarter of 2027 and reach full capacity by mid-2028. The agreement structure is worth noting: a capacity reservation ties the supplier's new lines to a named customer ahead of time, rather than waiting for open-market orders to arrive.
Advanced packaging: interposer wafers drive the greater than 300% plan
NVIDIA and Amkor intend to expand advanced packaging capacity in South Korea, targeting a greater than 300% increase in interposer wafer capacity by the end of 2028. The planned capacity expansion, together with Amkor Technology's development of new manufacturing facilities in South Korea, will support higher production volumes for NVIDIA's current and next-generation platforms. Packaging is the step that converts fabricated dies into usable accelerators, so a triple-digit jump here functions as the ceiling-raiser for everything downstream.
Testing: where the doubling comes from
NVIDIA expects its total semiconductor testing capacity to double from calendar 2026 to 2027, with buildouts in the U.S., Taiwan and Singapore providing support. The schedule is compressed: a full doubling in roughly one calendar year means the new sites must reach volume quickly, not merely open.
- King Yuan Electronics (KYEC): a $1.1 billion capacity reservation advance from NVIDIA will support new U.S. testing facilities and equipment, and high-volume production is targeted for April 2027.
- Singapore: expanded testing operations are expected to begin high-volume production in February 2027.
- US footprint: the effort also strengthens US manufacturing, adding a domestic node to a testing base previously concentrated in Asia.
Optical engines: co-packaged optics at 15x
NVIDIA and SPIL intend to expand manufacturing capacity in Taiwan for optical engines used in co-packaged optics, supporting the growing connectivity requirements of large-scale computing infrastructure. The expansion targets an increase of 15x by mid-2027, with aims for another doubling by the end of 2028. Taken together, the sequence implies roughly a thirtyfold step from today's baseline across the two-stage plan — the largest proportional move of any element in the package.
Capacity targets at a glance
| Capability | Partner and site | Commitment | Milestone |
|---|---|---|---|
| Substrates | IBIDEN, Ono Plant, Japan | multibillion-dollar capacity reservation advance; capacity reaching 3x current levels | additional capacity begins ramping in the first quarter of 2028 |
| Substrates | Samsung Electro-Mechanics, South Korea (Sejong, Busan) | immediate capacity from existing operations plus upgrades and new factories | a significant share of annual package substrate capacity By 2028 |
| Substrates | Unimicron Technology Corporation, Taiwan (KF) | capacity reservation agreement | adding capacity in the second quarter of 2027; full capacity by mid-2028 |
| Advanced packaging | Amkor, South Korea | greater than 300% more interposer wafer capacity | targeted by end of 2028 |
| Testing | King Yuan Electronics (KYEC) in the U.S.; Taiwan; Singapore | total testing capacity to double from calendar 2026 to 2027; $1.1 billion advance to KYEC | high-volume production in February 2027 (Singapore) and April 2027 (U.S.) |
| Optical engines | SPIL, Taiwan | co-packaged optics capacity to rise 15x | by mid-2027, with another doubling by the end of 2028 |
FAQ
What is the plan actually expanding?
Four upstream capabilities rather than chip fabrication itself: substrates, advanced packaging, testing and optical engines. These are the steps between a fabricated die and a shipping AI system.
How quickly will substrate capacity grow?
NVIDIA's stated goal is 2.5x for its products over the next three years. Supplier-level targets differ: IBIDEN's Ono Plant aims for capacity reaching 3x current levels, while Samsung Electro-Mechanics contributes immediate capacity from existing South Korean operations.
Which supplier receives the largest disclosed advance?
The figure disclosed for a single partner is the $1.1 billion capacity reservation advance to KYEC for new U.S. testing facilities and equipment. The IBIDEN support is described only as a multibillion-dollar advance, without a specific number.
What dates should readers hold the partners to?
February 2027 for high-volume testing output in Singapore; April 2027 for high-volume U.S. testing output; mid-2027 for the 15x optical engine target; the second quarter of 2027 for the start of Unimicron's added substrate capacity; the first quarter of 2028 for additional IBIDEN ramp; mid-2028 for full Unimicron capacity; and end of 2028 for both the packaging and the optical-engine follow-on targets.
Why does the geography matter?
The buildout spans Japan, South Korea, Taiwan, US and Singapore sites, and the plan explicitly strengthens US manufacturing. That distribution is described as broadening key manufacturing capabilities across new and existing suppliers, which strengthens the resiliency and redundancy of NVIDIA's global supply chain.
What the plan signals for the wider AI buildout
Two features of the disclosure stand out for anyone tracking infrastructure. The first is sequencing: testing capacity doubles first, from calendar 2026 to 2027, ahead of most substrate and packaging additions that land in 2028, which suggests validation throughput is being treated as the nearer-term constraint. The second is the concentration of risk in 2028 — the first quarter of 2028, mid-2028 and end of 2028 all carry major milestones, so the 2.5x substrate and greater than 300% packaging goals depend on several ramps arriving within the same twelve-month window.
For buyers and operators, the practical effect is a larger and more diversified pool of qualified suppliers for the least visible parts of the stack. The commitments themselves are still contingent: agreements are in progress, capacity reservations precede output, and the announced multiples describe intent until each site reaches high-volume production.