Nvidia Signs Six Suppliers to Expand Chip Substrate, Packaging, Testing and Optics Capacity

Nvidia is paying to lock up the unglamorous parts of the AI chip supply chain.

On Wednesday, Oct. 21, during CEO Jensen Huang's keynote on the second day of GTC Berlin 2026 at the Tempodrom, the company announced agreements with six manufacturers: Amkor Technology, IBIDEN, King Yuan Electronics (KYEC), Samsung Electro-Mechanics, Siliconware Precision Industries (SPIL) and Unimicron Technology.

Huang told the audience Nvidia is "in the midst of the largest infrastructure buildout in human history." He added: "Demand is extraordinary. Capacity is the constraint."

The targets

The agreements cover four layers: substrates, advanced packaging, chip testing and optical engines. Nvidia set these capacity goals:

• Substrates: 2.5 times current capacity within three years, through agreements with Samsung Electro-Mechanics, IBIDEN and Unimicron.
• Advanced packaging: an increase of more than 300 percent in interposer wafer capacity in South Korea by the end of 2028, with Amkor.
• Testing: total semiconductor testing capacity doubling from calendar 2026 to calendar 2027.
• Optical engines: a 15-fold production increase by mid-2027, for Nvidia's next-generation networking switches.

These are Nvidia's own targets, not delivered output.

Nvidia said it is committing a $1.1 billion capacity reservation advance to KYEC, the testing company. That is a prepayment: Nvidia puts up cash now to hold a supplier's future capacity.

What Samsung Electro-Mechanics gets

Samsung Electro-Mechanics, the components unit of Samsung Electronics, announced a strategic partnership with Nvidia on Oct. 21. The two companies will jointly invest in dedicated ABF package substrate lines at the Sejong and Busan plants in South Korea.

The supply covers substrates for Nvidia's compute chips and for the data center networking switches that link chips together. Under the multiyear agreement, deliveries begin in 2026, and volumes are expected to grow at a double-digit annual rate.

By 2028, production allocated to Nvidia is projected to be a significant portion of Samsung Electro-Mechanics' total annual substrate capacity, Nvidia said. The company will initially supply from existing Korean operations, while the partnership also supports facility upgrades and new factories.

Neither company's statement puts a dollar figure on the deal. KED Global reported, citing unnamed industry sources, that it is a multibillion-dollar contract and the largest the company has landed for FC-BGA substrates. That figure is unconfirmed.

Samsung Electro-Mechanics CEO Jang Deok-hyun said the company is "honored to join NVIDIA, a leader in global AI innovation, as a key infrastructure partner." Deborah Shoquist, Nvidia's executive vice president of operations, said "expanding production capacity on a global scale is necessary to keep pace with the speed of AI infrastructure development."

Samsung Electro-Mechanics had already said it would invest 23 trillion won, about $17.3 billion, through 2040 to develop the Sejong and Busan plants.

Why substrates matter

A package substrate is the multilayer board between a chip die and the circuit board. For a GPU-class AI accelerator, it has to handle extreme heat and route signals at sub-millimeter pitch across dozens of layers.

The key material is Ajinomoto Build-up Film, which Tech Times reports is made by a small number of manufacturers and cannot be substituted. That is why Nvidia cannot simply buy its way to more supply overnight, and why it is writing checks to suppliers rather than only placing orders.

Samsung Electro-Mechanics said larger substrates with embedded components raise manufacturing complexity. Warpage control and flatness are critical to yields. In its release, the company said it embeds passive components such as multilayer ceramic capacitors and silicon capacitors directly in the substrate.

The map

The supplier base spans Japan (IBIDEN), Taiwan (Unimicron, KYEC, SPIL), South Korea (Samsung Electro-Mechanics) and the United States (Amkor). The Amkor expansion Nvidia described is in South Korea, alongside new Amkor facilities there. The sites named for the substrate lines are also in South Korea.

Nvidia's announcement frames the deals as a bet that supply, not demand, limits its growth. The prepayments put Nvidia's own cash behind that view.

What comes next

The first measurable checkpoints are mid-2027 for the 15-fold optical-engine target and calendar 2027 for the doubling of testing capacity. The substrate and packaging goals run out to 2028. Nvidia has not published a schedule for interim progress, and Samsung Electro-Mechanics has not disclosed the contract's value.