Samsung Electronics Chairman Lee Jae-yong significantly strengthened the competitiveness and growth prospects of the company's core businesses — spanning memory chips, foundry and enterprise generative AI services — through his participation in the San Francisco AI Summit held in California.
According to industry sources Monday, Lee met in rapid succession with leading global AI players in Silicon Valley on Friday and Saturday (local time), including Nvidia, Broadcom, OpenAI and Anthropic, reinforcing Samsung's position in the memory chip business while exploring opportunities to expand its foundry operations.
Samsung is the only company in the world with integrated capabilities across memory, foundry and advanced packaging, and analysts say Lee made full use of that advantage. The meetings are expected to ease concerns about a peak in the memory cycle and lay the groundwork for long-term growth across Samsung's semiconductor business.
The centerpiece of the trip was a $200 billion partnership with fabless chipmaker Broadcom — a deal widely seen as a springboard for further collaboration with other big-tech companies.
Google, Amazon, Meta and Microsoft are all racing to develop their own custom application-specific integrated circuits (ASICs) as alternatives to Nvidia's GPUs, and Broadcom is the firm behind the design of those chips.
Big-tech companies have partnered with Broadcom to develop their own ASICs — chips that are faster, cheaper and more power-efficient for data training and inference — as they work to reduce dependence on Nvidia's expensive AI accelerators.
Broadcom has already secured the design contract for the seventh generation of Google's in-house AI chip, the Tensor Processing Unit (TPU), and has handled the design of Meta's MTIA chips through four generations, cementing its status as the de facto power broker in the ASIC market.
Samsung Electronics used the occasion to deepen its long-term partnership with Broadcom, agreeing to expand cooperation on next-generation AI chip development and jointly identify new business opportunities in the AI computing market.
The plan calls for supplying cutting-edge memory chips, including HBM, to enhance the performance of Broadcom-designed AI chips, while rapidly mass-producing and delivering AI chips built on sub-2-nanometer foundry process technology.
Samsung Electronics, which earlier won an order to produce Nvidia's new AI inference chip — the Groq3 LPU — on a 4-nanometer foundry process, is counting on its Broadcom partnership to secure additional orders at the 2-nanometer node. TSMC's production capacity constraints and its recent foundry price increases are seen as opening the door for Samsung's foundry business.
The quickening pace of activity by Han Jin-man, the president leading Samsung's foundry business, is adding to the optimism. Beyond attending the MOU signing ceremony with Broadcom, Han also appeared Sunday at "Advancing AI 2026," an event hosted by AMD — often described as Nvidia's closest rival — in San Francisco. AMD, which sources HBM from Samsung Electronics, is considered a key partner and has been in discussions with Samsung's foundry division about producing its own AI accelerators as well.
OpenAI and Anthropic, both of which Lee met during the trip, are also Samsung Electronics memory customers and potential foundry partners. OpenAI had already pledged a broad alliance covering both memory and foundry at a meeting between Lee and CEO Sam Altman in South Korea in October last year.
Anthropic, best known for its generative AI model Claude, is also in discussions with Samsung's foundry division about contract chip production based on a 2-nanometer process. The company is deepening its ties with Samsung more broadly — signing a strategic partnership with Samsung SDS to identify new AI business opportunities in South Korea.
joze@heraldcorp.com
