Two hundred billion dollars is a number that demands attention. Samsung Electronics and Broadcom have signed a memorandum of understanding at the AI Summit in San Francisco, committing to a broad collaboration across memory chips, foundry services, and advanced packaging through 2030. That’s not a product announcement. That’s a structural bet on who builds the backbone of AI infrastructure for the next half decade.
The deal covers three distinct areas. Samsung will supply high bandwidth memory for Broadcom’s next-generation AI accelerators, which puts it directly in competition with SK Hynix, the current HBM market leader and primary supplier to Nvidia. On the foundry side, Broadcom will use Samsung’s 2-nanometre and below process nodes for products including its Wireless Broadband Communications solutions. And the two companies plan to collaborate on advanced packaging, specifically 2.3D and 2.5D integration, to push performance and power efficiency for AI and networking chips.
That combination matters because AI chips are no longer just about raw compute. The bottleneck has shifted to how quickly data moves between memory and logic, and how efficiently multiple dies can be packaged together. TSMC has dominated that space, particularly through its CoWoS packaging technology used in Nvidia’s H100 and B100 series. Samsung is signaling it wants a serious slice of that work, and a committed customer like Broadcom gives it the volume and credibility to compete.
This deal fits a pattern Samsung has been building carefully. Reuters reported last month that Samsung’s Device Solutions CEO Young Hyun Jun held foundry talks with Nvidia CEO Jensen Huang, including discussions about HBM4E and HBM5 memory. Samsung also landed a $16.5 billion chip supply contract with Tesla and a separate agreement to manufacture an AI-enhanced modem for Starlink. Each deal adds weight to the next. Broadcom is not a minor win here. It’s one of the most important custom silicon players in the market, designing chips for hyperscalers like Google and Meta.
For developers and founders building on AI infrastructure, the downstream effect could be real. More foundry competition means more production capacity, which has been a genuine constraint on AI chip availability. And if Samsung closes the gap with TSMC on advanced packaging, hyperscalers gain more supplier options, which historically drives better pricing and faster timelines. So this is worth watching closely, even if the full impact won’t be visible until 2027 or later.
