Ibiden, the dominant supplier of chip package substrates used in Nvidia's cutting-edge semiconductors, may need to dial up the pace of production capacity increases to keep up with demand, according to its chief executive officer.

Sales of the 112-year-old company’s AI-use substrates are robust with customers buying up all that Ibiden can sell, CEO Koji Kawashima said in an interview, adding that that demand is likely to last at least through next year.

Ibiden is building a new substrate factory in Gifu Prefecture, expected to go online at 25% production capacity around the last quarter of 2025 before reaching 50% by March 2026. But that may not be enough, Kawashima said. The company is in talks about when to get the remaining 50% capacity online.

"Our customers have concerns,” he said. "We’re already being asked about our next investment and the next capacity expansion.”

Ibiden’s shares rose as much as 5.5% in Tokyo on Monday, their biggest intraday gain in more than a month.

Ibiden’s clients include Intel, AMD, Samsung and TSMC, as well as Nvidia. Many of them consult with the Japanese company early in product development, because the substrates — which help transmit signals from semiconductors to the circuit board — need to be tailored for each chip. Substrates must be made to withstand the heat of an Nvidia graphics processing unit to form an AI chip package complete with components such as memory.

Founded as a power utility company in 1912, Ibiden developed semiconductor expertise through a partnership with Intel that Kawashima cultivated by waiting every day in front of the Santa Clara company to stop engineers and executives for product feedback in the early 1990s. At one point, Intel comprised around 70% to 80% of Ibiden’s revenue from chip package substrates. That fell to around 30% in the fiscal year ended March as the US chipmaker struggled to execute a turnaround that recently saw the ousting of CEO Pat Gelsinger.

Reliance on Intel has hurt Ibiden’s stock, down around 40% this year. In October, Ibiden revised down its profit outlook after sluggish demand for components used in general purpose servers outweighed AI server-related growth. But while noting it was important to expand business with chipmakers other than Intel, Kawashima said he was confident Intel will bounce back.

"Intel’s overall technology is very sophisticated,” the 61-year-old said. "Intel raised us up and opened so many doors. Our relationship with Intel will always be our treasure, and Intel will forever be an important customer.”

With many foreign chipmakers unwilling to transfer their latest technology to the U.S., Intel is likely to play a key role in Washington’s goal to boost cutting-edge semiconductor production capabilities at home, Kawashima said. Ibiden has no manufacturing facilities in the U.S. itself. It has no plans to build any due to the cost of labor and logistics, Kawashima said, irrespective of U.S. President-elect Donald Trump’s plans to impose tariffs on a wide range of products.

All of Nvidia’s AI semiconductors now use Ibiden’s substrates, although Taiwanese rivals such as Unimicron are eyeing the field. But it won’t be easy to break Ibiden’s position as the dominant supplier, according to Toyo Securities analyst Hideki Yasuda.

"Nvidia’s AI chips need sophisticated substrates, and Ibiden is the only one that can mass-produce them at a good production yield,” he said. "Taiwanese competitors won’t be able to take Ibiden’s share away by much.”

AI semiconductors account for more than 15% of Ibiden’s sales of around ¥370 billion ($2.3 billion), with that percentage expected to rise further. Nvidia’s said it’s begun full production of its next-generation Blackwell chips after encountering some initial technical challenges.

Over the long term, Nvidia may face growing competition from application-specific chips by Marvell and Broadcom as well as in-house silicon from Google and Microsoft. In theory, Ibiden should be able to accommodate them all, as AI chip package design and material will likely remain similar to Nvidia’s, according to Kawashima.