AI supply expansion reaches packaging, memory contracts and factory-built data centers

NVIDIA’s October 21 agreements target packaging substrates, chip testing, advanced packaging and optical engines—separate manufacturing steps that can limit finished-system supply. It projects 2.5 times substrate capacity for its products over three years and more than 300% growth in Amkor interposer-wafer capacity by the end of 2028. These are supplier expansion targets, rather than delivered GPUs. [1]

Memory supply is also being contracted far ahead. On October 28, SK hynix said long-term agreements with 14 major customers cover more than half its total production capacity for 2027–28. Contracts typically span five years and adjust within agreed price bands. The company reported preliminary Q3 revenue of KRW104.6155 trillion and operating profit of KRW83.3338 trillion, an 80% reported operating margin. These consolidated K-IFRS figures remain subject to review. [2]

At the other end of the chain, NVIDIA’s DSX POD design moves integration into modular factories. The company says installation could fall from six months to as little as 14 days at a prepared site, with partner deliveries planned for early 2027. Power, cooling, networking and supporting infrastructure must already be ready for connection. That qualification defines what the proposed time saving actually covers. [3]