Samsung Electro-Mechanics AI Server Pivot: R&D Surges 36%, Capacity Cannot Match Big-Tech Demand

Shares of Samsung Electro-Mechanics crossed 2 million won intraday on Friday, May 29, doubling in twelve trading days as demand for the company's AI server components — flip-chip ball grid array (FCBGA) substrates and multilayer ceramic capacitors (MLCCs) — ran so far ahead of production capacity that the company's own CEO acknowledged it cannot keep up. The milestone arrived as the company's three-year R&D expansion, which lifted research spending by 36% between 2022 and 2025, reshaped a parts maker long dependent on smartphone components into one of the few global suppliers able to produce the substrates and capacitors that AI data centers require.

Samsung Electro-Mechanics Shifts from Smartphones to AI Servers

Samsung Electro-Mechanics — a Samsung Electronics affiliate that manufactures multilayer ceramic capacitors, camera modules, and semiconductor package substrates — has been systematically redirecting its research budget away from its legacy smartphone-component base and toward higher-margin parts for electric vehicles and AI servers. Research and development spending climbed from ₩577.1 billion ($387 million) in 2022 to ₩787.0 billion ($528 million) in 2025, an increase of approximately 36% over three years. As a share of revenue, R&D rose from 6.1% in 2022 to 7.0% in 2025 and reached 7.2% in the first quarter of 2026, when the company spent ₩231.8 billion ($155 million) on research alone.

The numbers reflect a deliberate portfolio pivot. On April 30, the company reported Q1 2026 revenue of ₩3.2091 trillion — the first quarter in its history to exceed ₩3 trillion — with operating profit up 40% year on year to ₩280.6 billion. The Package Solution division, which makes FCBGA substrates, posted a 45% revenue increase year on year, the steepest gain of any business unit. The Component Solution division, which makes MLCCs, grew 16% year on year.

Why FCBGA Substrates Matter for AI Server Builders

An FCBGA substrate is a high-density board that connects an advanced processor chip to the rest of a server system, routing signals and power between the chip's thousands of solder bumps and the mainboard below. As AI chips have grown larger and more complex — Nvidia's Blackwell GPU occupies roughly 2,700 square millimeters and stacks multiple dies — the substrates required have grown correspondingly difficult to produce. They now typically exceed 110 millimeters on a side, stack more than 20 layers of conductors and insulating film, and demand micron-level alignment tolerances across the entire area. Only a handful of companies worldwide can manufacture them reliably.

Samsung Electro-Mechanics became the first South Korean company to mass-produce server FCBGA substrates in October 2022. The company has since reorganized its substrate business around these higher-value products, secured first-vendor status for FC-BGA substrates on the Groq 3 language processing unit — an inference accelerator integrated into Nvidia's forthcoming Vera Rubin platform — and began mass production for that design in Q2 2026. In April, the company received an investment certificate from Vietnamese authorities for a $1.2 billion AI-related FCBGA production facility in Thai Nguyen.

"A structural shortage in FC-BGA for AI servers and data centers is set to continue this year," said Park Kang-ho, an analyst at Daishin Securities, who projected Samsung Electro-Mechanics could "leap to the global No. 1 position" in the segment through additional investment.

The competitor landscape underscores how much ground remains. Japan's Ibiden and Shinko Electric Industries together hold roughly 70% of the FCBGA substrate market. Samsung Electro-Mechanics and LG Innotek — both challengers — are expanding into the high-end segment as global data-center operators scramble for supply.

FCBGA Substrate Shortage: Why Orders Exceed Production

The immediate supply squeeze is severe. CEO Chang Duck-hyun told shareholders at the March annual general meeting that FCBGA demand exceeds the company's production capacity by more than 50%. Kim Yeon-mi, an analyst at Daol Investment and Securities, put the timeline in sharper relief: "Considering that meaningful capacity additions are not expected until the first quarter of 2028, demand is projected to exceed supply through 2027, making price hikes highly likely."

Samsung Electro-Mechanics has already begun raising prices on select FCBGA products, leveraging a seller's market. The company is also in discussions with big-tech data-center customers about long-term supply agreements for both FCBGA substrates and MLCCs, a step that would give both sides visibility into production planning over multiple quarters. The company has indicated it will more than double its facility investment in 2026 compared to the prior year and is reviewing additional capacity expansions from 2027 onward.

EV Capacitors and Autonomous-Vehicle Cameras

The same R&D push is flowing into automotive electronics. For electric vehicles, Samsung Electro-Mechanics has developed ultra-high-voltage MLCCs rated at 1,000 to 1,250 volts for EV charging systems — a range that addresses the industry shift toward higher-voltage architectures designed to reduce charging time and improve energy efficiency. The company has also built out a line of medium-voltage capacitors aimed at high-output fast-charging applications. Each additional level of autonomous-driving capability adds approximately 1,000 MLCCs per vehicle, providing a structural volume increase as ADAS adoption widens.

On the camera side, the company has completed a full lineup of high-resolution automotive camera modules for advanced driver-assistance systems, targeting the growing demand for cameras that can simultaneously deliver recognition accuracy, thermal durability, and mechanical reliability across vehicle operating conditions. The Q1 2026 earnings release confirmed that automotive substrate supply for ADAS and autonomous driving also expanded in the quarter.

Ko Eui-young, an analyst at iM Securities, said rising technical difficulty in MLCCs is cementing a duopoly between Samsung Electro-Mechanics and Japan's Murata Manufacturing in the AI server MLCC segment, a market structure that gives both companies significant pricing power as supply remains tight.

What Big-Tech Data-Center Buyers Need to Know

For procurement teams at hyperscaler and cloud operators, the supply picture at Samsung Electro-Mechanics has direct implications. The company's Q1 2026 earnings release noted that orders from big-tech data-center customers are surging, and that increasingly sophisticated substrates eat into production capacity, leaving the company unable to fully meet requested volumes with its current output. The company plans to respond aggressively to demand for high-value substrates tied to AI data centers and will continue to grow its FCBGA business.

The company's Q2 2026 guidance points to continued revenue growth: the Component Solution and Package Solution divisions both face demand running ahead of supply, and the company has flagged that new big-tech networking substrate customers will begin receiving full-scale shipments in Q2. NH Investment and Securities projects Samsung Electro-Mechanics revenue will reach ₩15.751 trillion by 2027, with operating profit surging from approximately ₩913 billion in 2025 to ₩3.007 trillion in 2027.

For procurement and engineering teams at AI server manufacturers, the practical window for securing 2027 substrate supply is now, before capacity expansions come online. Samsung Electro-Mechanics has signaled it will prioritize long-term agreement customers in allocating new production capacity from 2027.

Frequently Asked Questions

What is an FCBGA substrate, and why is it in short supply for AI servers?

A flip-chip ball grid array (FCBGA) substrate is a high-density board that connects advanced processor chips — such as AI accelerator GPUs and server CPUs — to the mainboard of a server system. As AI chips have grown larger and more power-hungry, the substrates required have grown more complex: they now measure over 110 millimeters on a side, carry more than 20 conductor layers, and demand micron-level manufacturing tolerances that only a small number of companies worldwide can achieve. Because building new FCBGA production lines takes multiple years, demand from AI data-center operators has far outpaced available supply.

How much has Samsung Electro-Mechanics increased its R&D spending on AI and EV components?

Samsung Electro-Mechanics grew its annual R&D spending by approximately 36% between 2022 and 2025, rising from ₩577.1 billion ($387 million) to ₩787.0 billion ($528 million). As a share of revenue, R&D increased from 6.1% to 7.0% over that period, reaching 7.2% in Q1 2026. The bulk of that investment has flowed into higher-margin components for AI servers and electric vehicles, including FCBGA substrates, high-voltage automotive MLCCs, and ADAS camera modules.

When will the FCBGA substrate shortage end?

Analysts do not expect meaningful new FCBGA production capacity to come online until early 2028 at the earliest. Kim Yeon-mi of Daol Investment and Securities said in late May 2026 that demand is projected to exceed supply through 2027, making further price increases likely. Samsung Electro-Mechanics has said it is reviewing investments to expand supply from 2027 onward, but has not committed to a specific capacity target or timeline.

What AI server components does Samsung Electro-Mechanics make?

Samsung Electro-Mechanics produces three categories of components critical to AI server infrastructure: FCBGA package substrates, which connect AI accelerator chips to server mainboards; multilayer ceramic capacitors (MLCCs), which stabilize power delivery inside AI servers; and, since 2025, silicon capacitors, which sit directly inside advanced semiconductor packages to manage power noise at the chip level. AI servers require three to four times as many MLCCs as conventional servers, and the technical demands on each component have risen sharply as AI chips have grown in size and power consumption.