NVIDIA Targets 2.5x Substrate Capacity, Doubles Testing by 2027

NVIDIA plans to multiply its substrate production capacity 2.5-fold within three years, the company announced on October 28, calling output capacity — not demand — the binding constraint on AI computing.

The announcement names six partners: Amkor Technology, Ibiden, King Yuan Electronics (KYEC), Samsung Electro-Mechanics, Siliconware Precision Industries (SPIL) and Unimicron. NVIDIA will support capacity expansion at both new and existing suppliers across substrates, advanced packaging, semiconductor testing and optical engines.

Founder and CEO Jensen Huang said: "The largest infrastructure buildout in human history is underway, and computing demand is surging, but the biggest constraint today is production capacity. We are grateful to our partners for working with us to scale production around the world and deliver AI supercomputing systems."

The claim is a plan, not shipped capacity. Every milestone cited carries a date between 2027 and 2028, so none of the announced volumes can be verified before then.

What does the substrate expansion involve?

• Ibiden will expand capacity at its Ono plant in Japan to roughly three times current levels, under an agreement announced early this year. NVIDIA is providing several billion dollars in capacity prepayments. The new capacity comes online in stages from the first quarter of 2028.
• Samsung Electro-Mechanics joins the substrate supply chain as a new entrant. The work covers upgrades to existing facilities in Sejong and Busan plus one new plant. By 2028, NVIDIA-allocated production is expected to take a significant share of the company's total annual package substrate capacity. NVIDIA did not quantify "significant."
• Unimicron will expand KF substrate (a high-density packaging board for AI chips) capacity in Taiwan, backed by a capacity reservation agreement. Capacity begins arriving in the second quarter of 2027 and reaches maximum output by mid-2028.

How far does advanced packaging scale?

NVIDIA and Amkor aim to raise interposer wafer (the silicon layer connecting a chip to its package) capacity by at least 300% by the end of 2028, in Korea. The expansion supports current and next-generation NVIDIA platforms alongside construction of Amkor's new Korean manufacturing facility.

Where does testing capacity double?

NVIDIA expects overall semiconductor testing capacity to double between 2026 and 2027, across the U.S., Taiwan and Singapore. A $1.1 billion capacity reservation prepayment to KYEC funds a new U.S. testing facility and equipment installation, with mass production targeted for April 2027. The expanded Singapore facility is scheduled to begin mass production in February 2027.

What about optical engines?

NVIDIA and SPIL will raise production of co-packaged optics (CPO, optical engines integrated directly with the switch chip) in Taiwan. The stated goal: 15 times current capacity by mid-2027, then double that level again by the end of 2028 — a 30-fold total increase to serve large-scale computing interconnects.

Three caveats sit inside the announcement. The dollar figure for Ibiden is "several billion," an unspecified range. The Samsung Electro-Mechanics capacity share is left undefined. And every date — February 2027 through end of 2028 — lies ahead, which makes today's numbers commitments rather than results.