NVIDIA details AI supply-chain capacity plans through 2028

NVIDIA outlined agreements and production plans with six suppliers to expand manufacturing capacity for AI infrastructure through 2028. The effort pairs signed capacity reservations and supply available now with production targets that have not yet been delivered.

For substrates, NVIDIA said agreements with IBIDEN, Samsung Electro-Mechanics and Unimicron are intended to raise aggregate capacity for its products by 2.5 times over the next three years. Its previously announced IBIDEN agreement includes a multibillion-dollar reservation advance for Japan’s Ono Plant. NVIDIA is targeting three times the plant’s current capacity, with additional output ramping in the first quarter of 2028. A separate reservation backs Unimicron’s KF expansion in Taiwan, where added capacity is expected from the second quarter of 2027 and full capacity is targeted for mid-2028.

Samsung Electro-Mechanics is the immediate addition to the substrate group. NVIDIA said capacity is available now from Samsung’s existing South Korean operations. Samsung said dedicated lines in Sejong and Busan will support multiyear supply beginning in 2026. The share of Samsung’s annual substrate capacity allocated to NVIDIA by 2028 remains a projection.

For advanced packaging, NVIDIA and Amkor plan to expand operations in South Korea, targeting an increase of more than 300% in interposer-wafer capacity by the end of 2028. Amkor disclosed a new $1.3 billion agreement for the South Korean expansion and said its multiyear NVIDIA agreements in Arizona and South Korea total $3 billion. Those funding commitments are separate from the future production target.

NVIDIA also expects its total semiconductor-testing capacity to double from calendar 2026 to 2027 through buildouts in the United States, Taiwan and Singapore. The company said a $1.1 billion capacity-reservation advance to KYEC will support new U.S. 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, though NVIDIA did not identify the operator or quantify each country’s contribution to the forecast.

For optical engines, NVIDIA and SPIL are targeting a 15-fold increase in Taiwan manufacturing capacity by mid-2027, then aim to double that expanded level again by the end of 2028. The engines support co-packaged optics, which places optical components close to switch silicon to shorten electrical paths and reduce power loss. ASE demonstrated that arrangement in 2025, while SPIL’s Tan Ke Plant had already reached the ramp-up stage for NVIDIA accelerated-computing demand that January.

The announcements did not disclose absolute unit capacity or consistent baseline measurements behind the multipliers. That means the targets cannot be converted into substrate, wafer, test or optical-engine volumes. NVIDIA classified the expansion amounts, production timing and expected benefits as forward-looking statements subject to risks and not guarantees of future performance.