According to Zhitong Finance App, U.S.-based semiconductor advanced packaging company Amkor Technology (AMKR.US) announced a major agreement worth up to $1.3 billion with "AI chip superpower" NVIDIA (NVDA.US) to expand its 2.5D/3D advanced packaging capacity in South Korea, after which the company's shares rose approximately 10% in pre-market trading on Wednesday.
Amkor stated in its latest press release: "With this latest commitment, Amkor Technology's multi-year agreements with NVIDIA in Arizona and South Korea total approximately $3 billion. These initiatives are designed to support the growing advanced packaging demand for next-generation AI and accelerated computing platforms."
These massive multi-billion-dollar investments aim to expand advanced chip packaging capacity to meet the growing computing resource demands from large tech giants in cloud computing and other sectors for high-performance AI chips and accelerated computing systems focused on AI data centers. Advanced packaging, compared to traditional packaging systems, helps integrate semiconductor components into a single package, thereby improving performance and energy efficiency.
NVIDIA Expands Capacity with Long-Term Advanced Packaging Agreements, Amkor Welcomes Rebound Catalyst
NVIDIA's capacity expansion plan announced at GTC Berlin is providing new fundamental support for previously under-pressure advanced packaging stocks. Amkor announced an additional $1.3 billion South Korea capacity expansion agreement with NVIDIA, bringing the total multi-year agreements in Arizona and South Korea to $3 billion. Related prepayments will support capacity construction and align with NVIDIA's next-generation AI platform requirements.
This latest news therefore drove its shares up approximately 10% in pre-market trading on October 21; previously, as of October 20, Amkor closed at $41.70, down 20.4% for the month of October, significantly underperforming the Philadelphia Semiconductor Index's approximately 10.6% decline over the same period. The market has thus gained concrete evidence to reassess its growth prospects: long-term capacity commitments from customers are providing demand support for U.S. domestic advanced packaging investments.
NVIDIA's advanced packaging partnership with the company has extended to the complete chip manufacturing and testing chain. Amkor signed a 10-year agreement in June this year with NVIDIA's largest chip foundry partner, TSMC, establishing a framework for TSMC to procure advanced packaging and testing services from Amkor; previously, both parties clarified that they would jointly determine the technologies adopted in Arizona, including Integrated Fan-Out (InFO) and CoWoS.
The core division of labor is to connect TSMC's advanced wafer manufacturing with Amkor's packaging and testing, enabling shared customers to obtain one-stop manufacturing services on U.S. soil. Because customer commitments exceeded the first phase's planned capacity, Amkor announced a second phase for the campus in September, raising the total planned investment to $12 billion, with the second phase expected to be completed by the end of 2029.
The More Complex the AI Agent Task Execution, the More Prominent the Value of Advanced Packaging
Amkor primarily engages in outsourced semiconductor assembly and testing (OSAT), covering markets including mobile phones, computing, automotive, and data centers; its capabilities most relevant to AI include 2.5D/3D through-silicon via (TSV), high-density fan-out (HDFO), Flip Chip, and heterogeneous integration. 2.5D packaging connects computing dies with high-bandwidth memory (HBM) through an interposer, while 3D integration further utilizes vertical interconnections to shorten data transmission paths; its S-SWIFT high-density fan-out platform connects multiple dies through fine redistribution layers (RDL).
Amkor's wafer thinning, interconnect processing, assembly, and testing are directly related to whether chips can be mass-produced and delivered under target bandwidth, power consumption, and reliability requirements. Agents like Astra and Muse combine multi-round reasoning, tool calling, and multimodal processing into continuously executing tasks; as the number, complexity, and concurrency of tasks increase, computing and memory access demands expand accordingly.
It follows that large cloud computing providers actively expanding their AI accelerator deployment scale will also continuously increase demand for high-density interconnections between computing dies and HBM, driving more complex packaging and testing investments. Amkor's growth space comes from both AI chip shipment expansion and increased packaging complexity per chip; NVIDIA's multi-year agreements and the cooperation framework with TSMC provide customer and manufacturing system support for Amkor to capture these incremental opportunities, and subsequent capacity utilization and yield ramp-up will determine the strength of revenue-to-profit conversion.
Comments