Samsung and Broadcom sign $200 billion AI chip pact

Samsung Electronics and Broadcom have signed a memorandum of understanding to expand their cooperation on chips for AI infrastructure. Announced on July 25 at an AI summit in San Francisco, the framework covers memory, contract manufacturing and advanced packaging. The companies estimate that the collaboration could exceed $200 billion over the five years through 2030. That figure describes the expected scale of the relationship, not one guaranteed purchase order.

Broadcom designs custom accelerators and connectivity chips used in large AI systems. Samsung plans to supply memory for Broadcom’s next-generation AI accelerators, including high-bandwidth memory, or HBM. HBM moves data quickly beside the processor, making it essential for demanding models. The foundry part of the agreement focuses on Samsung’s 2-nanometre and smaller manufacturing processes for Broadcom products, including high-speed wireless broadband communications chips. The companies also plan to work on 2.3D and 2.5D packaging, which combines several components inside one high-performance system.

The scope is important because the AI hardware race is no longer only about buying general-purpose GPUs. Large cloud providers and AI companies are increasingly commissioning specialised silicon for their own workloads. That shifts more value toward designers such as Broadcom and toward manufacturers that can provide logic, HBM and packaging together. It also makes memory supply, manufacturing yields and packaging capacity as important as the processor design itself.

For Samsung, a long-term Broadcom commitment could strengthen its foundry business as it tries to narrow the gap with TSMC. For Broadcom, closer access to Samsung’s integrated manufacturing stack offers another route to secure scarce components and turn custom designs into complete systems. The agreement therefore strengthens competition across a supply chain in which Nvidia, TSMC, SK Hynix and other major suppliers currently hold powerful positions.

AI users and software makers will not receive a new product immediately. The potential effect is indirect: more accelerator capacity, more competition and potentially better cost and energy efficiency in the data centres behind AI services. Companies planning infrastructure should remain cautious. The announcement gives no annual volumes, prices, named end customers or deployment timetable, and an MOU can change before orders are placed. Execution now depends on Samsung qualifying its HBM, achieving reliable yields on advanced processes and delivering complex packaging at scale. The headline number is striking, but those operational milestones will determine whether the pact actually changes the AI market.