Nvidia and Tesla to Fund $2.8 Billion Expansion of Samsung Electro-Mechanics' AI Chip Substrate Lines, Maeil Reports

Samsung Electro-Mechanics is set to receive 3.7 trillion won ($2.8 billion) in investment from Nvidia and Tesla to expand production of FC-BGA substrates used in AI semiconductors, according to an October 22 report by Maeil Business Newspaper, citing industry sources. The report was subsequently picked up and relayed by Reuters via Euronext. Neither Samsung Electro-Mechanics, Nvidia, nor Tesla is quoted directly in the available reporting confirming or denying the arrangement; the figures and structure described below rest on Maeil's sourcing.

What Maeil Reported

Per Maeil's account, Nvidia would put in 2.6 trillion won and Tesla 1.1 trillion won, a combined 3.7 trillion won (2.6 + 1.1 = 3.7) that matches the headline figure. The money would fund production lines in the South Korean city of Sejong, with substrates supplied under a long-term contract, the paper reported, citing industry sources.

The two customers' arrangements differ in scope. Nvidia's investment is described as covering expansion at both Samsung Electro-Mechanics' Sejong and Busan sites for AI semiconductor substrates. Tesla's commitment is narrower: a dedicated production line at the Sejong site specifically for substrates destined for autonomous vehicles and humanoid robots, according to Maeil.

Advance Payments, Not Equity

Maeil characterizes the funding as advance payment for equipment and facility expansion rather than an equity stake. The paper frames this as part of a broader industry pattern in which large technology buyers provide substrate manufacturers with equipment investment funds up front in exchange for stable supply over the long term. In Tesla's case, the roughly 1.1 trillion won is described as covering the funding needed to expand facilities and invest in equipment, securing supply for five years from 2029 to 2033 while reducing Samsung Electro-Mechanics' own investment burden, according to Maeil.

The reported deals sit alongside capacity commitments Nvidia has made publicly elsewhere. At GTC 2026 in Germany on October 21, Nvidia said it plans to raise production capacity for substrates used in its products to 2.5 times the current level over the next three years, and Maeil reported that it is also providing Japan's Ibiden with several billion dollars in advance payments and has signed an agreement with Taiwan's Unimicron to secure capacity. On the demand side, Samsung Electro-Mechanics CEO and president Jang Deok-hyun told the company's March shareholders' meeting that demand for FC-BGA used in servers and data centers runs at least 50% above production capacity.

Why capacity is hard to add quickly, as described in Maeil's reporting: AI accelerator substrates are several times larger than fingernail-sized mobile substrates and carry well over 20 circuit layers versus roughly 10, and must mount high-bandwidth memory and compute chips together, requiring narrower circuit lines and more precise layer alignment. Larger, more complex substrates are more prone to warping and microscopic defects, and even after new lines are built it takes considerable time to achieve stable yields. Our interpretation, tied to those reported constraints: funded capacity announced now would not translate into shipped volume immediately, which is consistent with the 2029-2033 supply window Maeil attributes to the Tesla arrangement. Maeil also reported that Samsung Electro-Mechanics is positioning its ability to supply FC-BGA alongside multilayer ceramic capacitors and silicon capacitors as a turnkey package, and is pursuing technology to embed silicon capacitors inside substrates.

Bottom line: every figure above, including the 3.7 trillion won total, the Nvidia and Tesla split, the site locations and the 2029-2033 supply window, rests on Maeil Business Newspaper's October 22 reporting and its industry sources, relayed by Reuters. Samsung Electro-Mechanics, Nvidia and Tesla have not publicly confirmed or denied the arrangement in the available reporting, and readers should treat the terms as reported rather than announced.