Zhitong Finance App learned that Amkor Technology (AMKR.US), an advanced semiconductor packaging company headquartered in the US, announced that it had reached an important agreement of up to 1.3 billion US dollars with NVDA.US (NVDA.US), the “AI chip superleader” to expand its 2.5D/3D advanced packaging production capacity in South Korea. The company's stock price then rose about 10% in pre-market trading on Wednesday.
Amkor said in a recent press release: “With this latest commitment, the total multi-year agreement between Amkor Technology and Nvidia in Arizona and South Korea has reached approximately $3 billion. These initiatives are aimed at supporting the growing demand for advanced packaging for next-generation AI and accelerated computing platforms.”
These large-scale multi-billion dollar investments are aimed at expanding the production capacity of advanced chip packaging to meet the growing demand for computing power resources from large technology giants such as cloud computing for high-performance AI chips and accelerated computing systems focused on AI data centers.
Compared to traditional packaging systems, advanced packaging helps integrate semiconductor components into a single package, thereby improving performance and energy efficiency.
Nvidia expands production with a long-term advanced packaging agreement, and Amkor ushered in a backlash catalyst
Nvidia's Berlin GTC production expansion plan is providing new fundamental support for advanced packaging stocks that were previously under pressure. Amkor announced an additional 1.3 billion US dollars to expand production in Korea with Nvidia. The total multi-year agreement between the two sides in Arizona and South Korea reached 3 billion US dollars. The relevant advance payments will support production capacity construction and connect with Nvidia's next-generation AI platform needs.
As a result, this latest news contributed to a pre-market rise of about 10% on October 21; previously, as of the 20th, Amkor closed at $41.70, with a cumulative decline of 20.4% in October, which is significantly weaker than the 10.6% decline in the Philadelphia Semiconductor Index during the same period. As a result, the market received a concrete basis for reevaluating its growth prospects: customers' long-term capacity commitments are supporting demand for advanced packaging investments in the US.
Nvidia's advanced packaging partnership with the company has extended to the complete chip manufacturing and testing chain. Amkor and TSMC, Nvidia's largest chip foundry partner, signed a 10-year agreement in June this year to establish a framework for TSMC to procure advanced packaging and testing services; previously, the two sides made it clear and jointly determined the technologies to be adopted in Arizona, including integrated fan-out packaging (InFO) and CoOS. Its core division of labor links TSMC's advanced wafer manufacturing with Amkor's packaging testing, so that joint customers can obtain one-stop manufacturing services in the US. As the customer promised to exceed the planned capacity of Phase I, Amkor announced Phase II of the park in September, increasing the total planned investment of the project to US$12 billion. The second phase is expected to be completed by the end of 2029.
The more complex the task execution of AI agents, the more prominent the value of advanced packaging
Amkor specializes in outsourced semiconductor packaging and testing (OSAT), covering markets such as mobile phones, computing, automobiles, and data centers; the most relevant capabilities for AI include 2.5D/3D silicon through-hole (TSV), high-density fan-out (HDFO), flip chip (Flip Chip), and heterogeneous integration. The 2.5D package connects the computing chip to the high-bandwidth memory (HBM) through an intermediary layer, and the 3D integration further uses vertical interconnection to shorten the data transmission path; its S-SWIFT high-density fan-out platform connects multiple grains through a fine heavy wiring layer (RDL). Amkor's wafer thinning, interconnection processing, assembly and testing are directly related to whether the chip can be mass-produced and delivered under target bandwidth, power consumption, and reliability requirements.
Agents such as Astra and Muse combine multiple rounds of inference, tool calls, and multi-modal processing into continuously executed tasks; as the number, complexity, and concurrency of tasks increase, computing and memory access requirements expand accordingly.
It is deduced from this that large cloud computing vendors are actively expanding the deployment scale of AI accelerators, and will also continue to increase the demand for high-density interconnections between computing chips and HBM, driving investment in more complex packaging and testing. Amkor's growth space also comes from the expansion of AI chip shipments and the increase in the complexity of single chip packaging; Nvidia's multi-year agreement and TSMC's cooperation framework provide customer and manufacturing system support to undertake these increases. The subsequent rise in capacity utilization and yield will determine the strength of revenue to profit conversion.