Nvidia Pays Up Front: $1.1 Billion to KYEC and a Multi-Billion Deal With IBIDEN
The announcement at GTC Berlin on October 21, 2026 comes down to one mechanism: Nvidia puts money up front so its partners will build factories and test facilities. The capacity figures — 2.5x substrates, more than 300 percent more interposer wafers, 15x optical engines — are, for now, the company's own targets, not documented capacity.
The News
At GTC Berlin on October 21, 2026, Nvidia announced plans with six supply-chain partners to expand global manufacturing capacity for the company's accelerated computing and full-stack platform for AI factories (Nvidia's press release via GlobeNewswire). The partners are Amkor, IBIDEN, KYEC, Samsung Electro-Mechanics, SPIL and Unimicron, and the areas of focus are four: substrates, advanced packaging, testing and optical engines.
The concrete element of the announcement is a set of dated milestones running from February 2027 to the end of 2028 — a timeline that provides a concrete basis for tracking whether the capacity actually materializes. The underlying idea, laid out in the press release by Nvidia founder and CEO Jensen Huang: "We are in the midst of the largest infrastructure buildout in human history. Demand is extraordinary. Capacity is the constraint." In other words: demand is not the bottleneck — capacity is.
The Mechanism: Capacity Reservation as Prepayment
The announcement's central instrument is what Nvidia calls "capacity reservation advances." Instead of waiting for suppliers to expand on their own and at their own expense, Nvidia puts capital up front so that the partner fabs get built. This ties the suppliers' production to Nvidia while shifting the investment burden onto Nvidia.
Most of the financial terms have not been disclosed. Two amounts are known: $1.1 billion to KYEC, and an advance of several billion dollars to IBIDEN without an exact figure. For the Unimicron, SPIL, Amkor and Samsung Electro-Mechanics agreements, neither amounts nor terms have been made available.
Partner by Partner
IBIDEN. Under an agreement IBIDEN signed and announced earlier this year, Nvidia is providing an advance of several billion dollars for the Ono plant in Japan. The expansion is intended to bring capacity to three times current levels, with ramp-up of additional capacity beginning in the first quarter of 2028.
Unimicron. The expansion of KF substrate production in Taiwan is scheduled to begin contributing capacity in the second quarter of 2027 and to reach full capacity by mid-2028.
Samsung Electro-Mechanics. The company is new to Nvidia's substrate supply chain and contributes immediate capacity from existing operations in South Korea. The collaboration also supports facility upgrades and new factories in Sejong and Busan. By 2028, production allocated to Nvidia is, according to the announcement, estimated to account for a substantial share of the company's total annual capacity for package substrates — with no specific figure (Unite.AI, which reproduces the same press release).
Amkor. Nvidia and Amkor plan to expand advanced packaging capacity in South Korea, targeting an increase in interposer wafer capacity of more than 300 percent by the end of 2028.
KYEC. Nvidia's $1.1 billion advance is to support new test facilities and equipment in the United States, with large-scale production targeted for April 2027. Expanded testing in Singapore is scheduled to begin large-scale production in February 2027.
SPIL. The goal is to increase capacity for optical engines for co-packaged optics fifteenfold by mid-2027, with a further doubling targeted by the end of 2028.
Taken together, Nvidia expects its total semiconductor test capacity to double from calendar year 2026 to 2027, supported by expansion in the United States, Taiwan and Singapore. The substrate agreements with IBIDEN, Samsung Electro-Mechanics and Unimicron are intended, in sum, to deliver a capacity increase of 2.5 times over the next three years.
What Is Verified — and What Is a Forecast
All capacity figures in this story come from Nvidia's own announcement. That applies to the 2.5x for substrates, the more than 300 percent for interposer wafers, the 15x for optical engines, and the doubling of test capacity. None of the figures has been independently confirmed, and there are no partner-side confirmations in the available source material. Unite.AI reproduces the announcement, including the Samsung Electro-Mechanics portion, and adds no independent verification.
Two caveats are worth highlighting. First, the financial terms of most of the agreements have not been disclosed, so it is impossible to assess the actual size of the investment beyond the KYEC figure and IBIDEN's "several billion dollars" designation. Second, the announcement uses vague phrasing on key points — a "substantial share" of Samsung Electro-Mechanics' capacity with no number. These are targets, then, not completed capacity.
What to Watch
The announcement contains dated checkpoints that make it possible to verify whether the targets are being met:
- February 2027: Singapore testing is scheduled to begin large-scale production.
- April 2027: KYEC's new test facilities in the US are to reach large-scale production.
- Second quarter 2027: Unimicron's KF expansion in Taiwan is to begin delivering capacity; SPIL's 15x target for optical engines is to be met by mid-2027.
- First quarter 2028: IBIDEN's Ono plant is to begin ramping toward tripled capacity.
- End of 2028: Amkor's target of more than 300 percent growth in interposer wafers, SPIL's further doubling, and Samsung Electro-Mechanics' "substantial share" are to be fulfilled.
Each of these dates is a concrete checkpoint. If the February 2027 Singapore milestone slips, that is an early signal that the timeline is optimistic. If they are met, Nvidia will have documented that the advance-financing model actually builds capacity — not just reserves it on paper.