NVIDIA plans 2.5 times the substrate capacity for its products over three years

Key points

• On 21 October 2026 NVIDIA announced plans with six supply-chain partners, Amkor, Ibiden, KYEC, Samsung Electro-Mechanics, SPIL and Unimicron, to expand capacity in substrates, advanced packaging, testing and optical engines.^[1]
• NVIDIA plans to increase substrate capacity for its products by 2.5 times over the next three years. Its agreements with Ibiden, Samsung Electro-Mechanics and Unimicron contribute to the target, and more are under way.^[1]
• The release dates the substrate steps as targets. Unimicron's expansion in Taiwan is to begin adding capacity in the second quarter of 2027 and reach full capacity by mid-2028; added capacity at Ibiden's Ono plant in Japan is to begin ramping in the first quarter of 2028; and by 2028 a significant portion of Samsung Electro-Mechanics' annual package-substrate capacity is projected to go to NVIDIA.^[1]
• Ibiden's board approved about ¥280 billion for the Ono plant in February 2026, with phased operation from fiscal 2027, and a further investment of about ¥151 billion for its Gama plant on 29 October 2026, with phased operation from fiscal 2029.^[6][8] Samsung Electro-Mechanics plans mass production from a KRW 4.27 trillion expansion at Sejong from September 2028.^[3]
• An IC package substrate is part of the chip package and carries signals between the chip and the main board, which is the printed circuit board.^[4] The capacity in these plans is substrate capacity and is reported here apart from board capacity. Substrates and AI boards draw on specialty glass cloth in different grades, and the low-CTE cloth for high-end substrates is reported as tight.^[10][11]

Context

On 21 October 2026 NVIDIA announced plans with a group of supply-chain partners to expand manufacturing capacity for its accelerated computing and AI factory platform. The partners are Amkor Technology, Ibiden, King Yuan Electronics (KYEC), Samsung Electro-Mechanics, Siliconware Precision Industries (SPIL) and Unimicron Technology. The plans cover four kinds of capacity: substrates, advanced packaging, testing and optical engines.^[1]

The release gives most of its figures as multiples of current capacity, reached by a stated date. They are targets and expectations, and this article reports each one with the measure it applies to. Samsung Electro-Mechanics and Ibiden have published their own statements and filings on their capacity plans, and these are set beside the release.^[2][3][6][7]

Board plants in Thailand, with their owners and stages, are listed in the MPBxChange plant register and followed in the tracker. The largest board makers and where they make boards are set out in the 2025 ranking, and the market as a whole in the world PCB market in 2026.

The partners and the dated steps

The release names a partner, a place and, for most steps, a period. The table sets out each partner's part as NVIDIA states it.

Substrates: Ibiden, Samsung Electro-Mechanics and Unimicron

NVIDIA plans to increase substrate capacity for its products by 2.5 times over the next three years. The release says the agreements with Ibiden, Samsung Electro-Mechanics and Unimicron will, in aggregate, drive this increase, and that more are under way.^[1]

Under an agreement that Ibiden signed and announced earlier in 2026, NVIDIA is providing a multibillion-dollar capacity reservation advance to support Ibiden's Ono plant in Japan. The expansion targets three times the plant's current capacity, with added capacity expected to begin ramping in the first quarter of 2028.^[1]

Samsung Electro-Mechanics joins NVIDIA's substrate supply chain with immediate capacity from its existing operations in South Korea, and the collaboration supports facility upgrades and new factories in Sejong and Busan.^[1] In its own release of the same day, Samsung Electro-Mechanics describes a strategic partnership under which the two companies will jointly invest in dedicated facilities for ABF package substrates used in AI semiconductors, with dedicated lines at Sejong and Busan. The supply covers package substrates for NVIDIA's compute chipsets and for high-speed networking switches. Deliveries start in 2026, and the volume is expected to grow at a double-digit annual rate over several years.^[2]

On 29 September 2026 Samsung Electro-Mechanics announced KRW 4.27 trillion to expand FCBGA production at its Sejong site, with investment from September 2026 to May 2028 and mass production from September 2028. The same release sets out about KRW 2.51 trillion for its FCBGA plant in Vietnam, which aims to begin operations by June 2030. The company said the investment includes customer funding support and mid- to long-term volume guarantees; the release does not name the customers.^[3]

NVIDIA supports Unimicron's KF substrate manufacturing expansion in Taiwan with a capacity reservation agreement. The expansion is expected to begin adding capacity in the second quarter of 2027 and to reach full capacity by mid-2028. The release gives no amount for this agreement or for the one with Samsung Electro-Mechanics.^[1]

Advanced packaging, testing and optical engines

NVIDIA and Amkor plan to expand advanced packaging capacity in South Korea, with a target of more than a 300% increase in interposer wafer capacity by the end of 2028, alongside new Amkor facilities in South Korea.^[1]

NVIDIA expects its total semiconductor testing capacity to double from calendar 2026 to 2027, with buildouts in the United States, Taiwan and Singapore. Its US$1.1 billion capacity reservation advance to KYEC supports new US testing facilities and equipment, with high-volume production targeted for April 2027. Expanded testing operations in Singapore are expected to begin high-volume production in February 2027.^[1]

With SPIL, NVIDIA plans to expand capacity in Taiwan for optical engines used in co-packaged optics. The target is an increase of 15 times by mid-2027, with another doubling aimed for by the end of 2028.^[1]

Stated targets against different starting levels

Each multiple in the release applies to its own measure. The 2.5 times applies to substrate capacity for NVIDIA's products, the 3 times to Ibiden's Ono plant, the increase of more than 300% to Amkor's interposer wafer capacity, the doubling to NVIDIA's total semiconductor testing capacity, and the 15 times to SPIL's optical-engine capacity. The release publishes none of the starting levels.^[1] The multiples therefore describe growth against different bases and cannot be added into one volume.

An increase of more than 300% means more than four times the starting level, and the chart draws Amkor's target at four times. All five multiples are targets for capacity; none describes output to date.^[1]

Ibiden gives a measure of its own. In February 2026 it said that its SAP (semi-additive process) capacity for AI and ASIC products, taking the end of the first half of fiscal 2024 as 1.0, is expected to reach just under 3.0 times by the end of fiscal 2028.^[7] That measure has a different scope and base from the 3 times for the Ono plant in NVIDIA's release.^[1]

Ibiden: the Ono plan and the decision of 29 October

Ibiden's notices of 2026 set out three capacity decisions for high-performance IC package substrates.

On 3 February 2026 Ibiden's board approved an investment plan of about ¥500 billion for its electronics business over fiscal 2026 to 2028. Its first phase, about ¥220 billion, centres on the existing Gama plant (Cell6) in Ogaki, Gifu Prefecture.^[5] On 26 February the board approved an investment of about ¥280 billion centred on the Ono plant (Cell8), to raise capacity for high-performance IC package substrates mainly for AI servers; the two phases make up the plan of about ¥500 billion.^[6]

In its summary of analysts' questions published on 27 February, Ibiden said that demand for AI servers was strong and that, as substrates gain layers and grow larger, its SAP capacity for leading-edge products would fall short. It said it had negotiated business terms with the customer and agreed on the direction before the board decision, and that its basic policy is to sign contracts with customers and receive the equivalent of the investment as prepayment.^[7] Ibiden's documents do not name the customer. NVIDIA's release refers to an agreement that Ibiden signed and announced earlier in 2026, under which NVIDIA provides a multibillion-dollar capacity reservation advance for the Ono plant.^[1] The two date the new capacity in different calendars: NVIDIA by calendar quarter, Ibiden by fiscal year.^[1][6]

On 29 October 2026 the board approved a further investment of about ¥151 billion centred on the Gama plant's Cell6 building and other overseas plants, for high-performance IC package substrates for CPUs, with phased operation and mass production from fiscal 2029.^[8] Ibiden told analysts on 30 October that start-up is planned for the second half of fiscal 2029, and that it expects a new factory building at its Gama or Ono plant to take at least four years from the decision.^[9] The decision is also set out in the article on the supply events of 27 to 29 October.

Ibiden also described the market for leading-edge AI substrates on 30 October. It said that a Korean maker is joining, making a field of three suppliers, and that the capacity of all competitors combined does not reach total demand. Taking customers' demand information into account, it described capacity as overwhelmingly short.^[9]

A package substrate and the board beneath it

An IC package substrate is part of the chip package. Samsung Electro-Mechanics describes package substrates as high-density circuit boards, with more microcircuits than general substrates, that carry electrical signals between the semiconductor and the main board.^[4] In AI servers the substrate is an FCBGA, which connects AI accelerators, GPUs and CPUs to the mainboard and supplies them with signals and power.^[3] The main board is the printed circuit board on which the finished package is mounted. The two layers are drawn in cross-section in the material shortage article, and the board itself is explained in what a PCB is.

The capacity in NVIDIA's plan, in Ibiden's notices and in Samsung Electro-Mechanics' Sejong investment is capacity for package substrates.^{[1][3][5][6][8]} It is reported here apart from the capacity of printed circuit board plants. The Taiwan Printed Circuit Association reports the substrate market as a figure of its own: world substrate output of US$14.75 billion in 2025, up 18.5%, and an estimated US$19.51 billion in 2026, up 32.3%, driven by AI servers, high-speed switches, ASICs, GPUs and high-end memory.^[10]

Substrates and the board chain

Substrates and AI boards both use glass cloth as reinforcement, in different grades. Low-CTE glass cloth is used mainly in IC substrates, and low-Dk cloth in the board layers beneath AI chips; T-glass, a low-CTE cloth, is a critical material for ABF and BT substrates.^[11] The Taiwan Printed Circuit Association reports that supply of the low-CTE glass cloth needed for high-end substrates remains tight, that new capacity needs time to start and ramp up, and that new suppliers must pass customer certification, so supply is hard to raise markedly in the short term.^[10] Ibiden said in February that the materials for its guidance for the current and next fiscal years were largely secured, and that it was qualifying new suppliers for volumes above that guidance.^[7] The supply of glass cloth, copper foil and laminate for boards is set out in the material shortage article.

The packages are mounted on boards that are themselves in demand. Victory Giant Technology, a Chinese board maker, stated on 30 October 2026 that, as AI computing infrastructure expands, demand for AI servers and for the high-end HDI and high-layer-count boards used with GPUs is growing fast.^[12] Its investment plans are covered in the Victory Giant article, and board demand by product in the world PCB market in 2026.

The release's next dated steps run from February 2027 to the end of 2028: high-volume testing in Singapore in February 2027 and at KYEC's new US facilities in April 2027; Unimicron's first added capacity in the second quarter of 2027; SPIL's 15 times by mid-2027; the start of the Ono ramp in the first quarter of 2028; Unimicron at full capacity by mid-2028; and Amkor's interposer target and SPIL's further doubling by the end of 2028.^[1] Samsung Electro-Mechanics plans mass production from its Sejong expansion in September 2028,^[3] and Ibiden plans phased operation of its Gama investment of 29 October from fiscal 2029.^[8] Events at board plants in Thailand are logged with their sources in the tracker, and the plants are listed in the register.

Data and method

This article draws on NVIDIA's release of 21 October 2026; Samsung Electro-Mechanics' releases of 29 September and 21 October 2026 and its package-substrate product page; Ibiden's timely disclosures of 3 February, 26 February and 29 October 2026 and its summaries of analysts' questions and answers of 27 February and 30 October 2026; the Taiwan Printed Circuit Association's release of 10 September 2026; TrendForce (24 November 2025); and Victory Giant Technology's announcement 2026-083 to the Shenzhen Stock Exchange of 30 October 2026. Each quotation was read on the publisher's page or in the filing on 1 November 2026.

Steps are dated as the sources date them: by calendar month, quarter, half or year, and for Ibiden by fiscal year. Multiples, capacity steps, investment amounts and start dates are the companies' targets and plans and are not counted as output. Unimicron's, Amkor's, KYEC's and SPIL's parts are taken from NVIDIA's release. Amounts are given in each company's own currency and are not converted. IC package substrates are reported apart from printed circuit boards. No company was contacted. Interpretation is MPBxChange analysis.