Samsung Electronics and Broadcom recently announced a long-term, deep cooperation agreement, pushing the strategic partnership between the two companies in the semiconductor field to a new height. This agreement covers the entire chain of memory chips, foundry manufacturing, and advanced packaging, and is expected to break through $200 billion in business collaboration by 2030, marking that the AI chip industry is entering a new stage of "design + manufacturing" deep integration.

Samsung 2

Sub-2nm process technology comes into place, joint efforts on HBM

In terms of technology, the core of the cooperation lies in fully integrating Broadcom's ASIC custom chip design capabilities with Samsung's advanced manufacturing processes. Broadcom's next-generation communication chips will be produced using Samsung's sub-2nm process, which means that chip scaling technology has reached the frontier of physical limits, and power efficiency and thermal management face severe challenges.

The two companies are also jointly developing the next generation of high-bandwidth memory (HBM). As AI model parameters continue to grow, memory bandwidth has become a key bottleneck for performance, making close collaboration between logic chips and memory chips more critical than ever before. This cooperation clearly reflects the changing direction of the entire industry: tech giants are moving away from relying solely on general-purpose GPUs toward developing customized chips for specific workloads, thus driving the chip design and manufacturing segments to unprecedentedly tight integration.

For Samsung, locking in a top client like Broadcom not only significantly strengthens its competitive position against TSMC in the foundry market, but also enhances the capacity utilization of its most advanced factories. Although Samsung previously had a $1.65 billion chip contract with Tesla, this agreement spans the entire manufacturing process, rather than being limited to a single product line. Previously, Samsung had already announced that it would secure more orders on the 2nm process and continue to advance next-generation memory collaborations such as HBM4E and HBM5.

As the demand for AI infrastructure becomes increasingly specialized, the collaborative efficiency of chips, memory, and packaging has become the core engine determining the upper limit of hardware performance — and Samsung and Broadcom clearly have already bet all their chips on this track.