Our Take, Up Front
The bottleneck in the artificial intelligence buildout is not demand, and it is not software. It is manufacturing capacity. Nvidia understands this better than anyone, which is why it is now spending tens of billions of dollars across its supply chain to lock in substrate, packaging, testing, and optical capacity years ahead. The scale of that effort was laid out on October 21 at GTC Berlin, and it is the clearest signal yet that the AI infrastructure race has moved from chip design to industrial capacity.
Nvidia announced capacity agreements with six manufacturing partners: Amkor Technology, IBIDEN, King Yuan Electronics, Samsung Electro-Mechanics, Siliconware Precision Industries, and Unimicron Technology Corporation. The headline targets are striking. Substrate capacity for Nvidia products is expected to rise 2.5 times over the next three years. Amkor plans to expand advanced packaging capacity in South Korea by more than 300 percent by the end of 2028. Nvidia total semiconductor testing capacity is expected to double from calendar 2026 to 2027 across the United States, Taiwan, and Singapore. And SPIL is targeting a 15-fold increase in optical-engine capacity for co-packaged optics by mid-2027, with plans for another doubling by the end of 2028.
Read the individual agreements and the strategy becomes clear. With IBIDEN, Nvidia is providing a multibillion-dollar capacity reservation advance to support the Ono plant in Japan, targeting three times current capacity with additional output beginning to ramp in the first quarter of 2028. Samsung Electro-Mechanics is joining the substrate supply chain with immediate capacity from existing operations in South Korea, plus new factories in Sejong and Busan; by 2028, production allocated to Nvidia is projected to account for a significant portion of the company total annual package substrate output. With Unimicron in Taiwan, a capacity reservation agreement supports the KF substrate expansion, with new capacity from the second quarter of 2027 and full capacity by mid-2028. For testing, Nvidia is advancing $1.1 billion to King Yuan Electronics for new United States facilities and equipment, targeting high-volume production in April 2027, while expanded testing operations in Singapore are expected to begin high-volume production in February 2027.
Jensen Huang, founder and chief executive of Nvidia, framed the moment in characteristically direct terms: the world is in the midst of the largest infrastructure buildout in human history, demand is extraordinary, and capacity is the constraint. The partners, he said, are the best manufacturers in the world, and the capacity they are adding is what will deliver the AI supercomputing systems powering the next industrial revolution.
Our view is that this announcement deserves to be read as a financial and strategic template, not just a manufacturing update. Capacity reservation advances, in which a customer pays a supplier billions of dollars today to secure output years from now, are a form of forward contracting that spreads risk and locks supply. Nvidia is effectively financing its suppliers expansion in exchange for guaranteed allocation, which is a decisive advantage in a market where every major cloud provider is competing for the same wafers, interposers, and test capacity. The geography matters too: capacity is being added in Japan, South Korea, Taiwan, Singapore, and the United States, a deliberate attempt to build redundancy across regions rather than concentrating production in any single point.
The demand side of the equation explains why Nvidia is willing to spend at this scale. Amazon Web Services announced in August that it plans to deploy 2 million additional Nvidia GPUs across its global infrastructure in 2027 and 2028, on top of earlier commitments exceeding 1 million GPUs starting in 2026. Nvidia and SK hynix announced a multiyear technology partnership in June to codevelop next-generation memory aligned to the AI infrastructure roadmap. The hyperscalers are spending at unprecedented rates: Meta raised its 2026 capital expenditure guidance floor to $135 billion with a cap of $145 billion, and Alphabet guided 2026 capital spending to between $195 billion and $205 billion, with spending expected to exceed $300 billion in 2027. Every one of those dollars ultimately flows through Nvidia supply chain in some form.
There are, of course, risks. Capacity buildouts of this size carry execution risk, and demand for AI infrastructure, however strong today, has never been tested across a full economic cycle at this scale. The reservation model also concentrates leverage: if demand softens, Nvidia carries large forward commitments to suppliers. And the political dimension is real, as data center siting, energy, and trade policy become contested issues in the United States and Europe.
But for B2B buyers, equipment suppliers, and logistics partners across the AI value chain, the practical conclusion is straightforward. Capacity, not demand, is the binding constraint, and the companies that secure capacity early will win the AI era. Nvidia has now committed the balance sheet to make sure it is one of them. Suppliers with the scale to meet these requirements will enjoy years of predictable demand; those without it will watch the buildout pass them by.
