Samsung Electronics and Broadcom have signed a memorandum of understanding (MOU) valued at more than $200 billion, marking one of the largest semiconductor partnerships ever disclosed. The agreement, announced on Friday at an AI summit in San Francisco, spans memory chips, foundry manufacturing, and advanced packaging for artificial intelligence (AI) semiconductors. It runs for five years through 2030 and covers some of the most critical layers of the AI chip supply chain. While the MOU represents a statement of intent rather than a binding contract, the scale of the commitment signals a major strategic alignment between two industry giants.
Memory: Next-generation HBM for AI accelerators
On the memory side, Samsung will supply Broadcom with next-generation high-bandwidth memory (HBM), including HBM4 and HBM4E chips, for use in Broadcom’s AI accelerators. High-bandwidth memory is a critical component for AI workloads because it provides the high data throughput required to feed increasingly powerful processors. HBM4, expected to enter volume production in 2025, will offer significant improvements in bandwidth, power efficiency, and capacity compared to current HBM3 and HBM3E generations. Samsung has been a leading player in the HBM market, competing closely with SK Hynix, which supplies Nvidia’s AI GPUs. By securing Broadcom as a customer for HBM4 and HBM4E, Samsung diversifies its AI memory revenue beyond the current concentration around a few hyperscalers.
Foundry: 2nm and smaller process technologies
In foundry, Samsung will manufacture Broadcom products using two-nanometer (2nm) and smaller process technologies at its campus in Pyeongtaek, South Korea. Samsung’s foundry business has long played second fiddle to TSMC, which controls more than 90 percent of leading-edge chip production. Samsung has struggled with yield rates on its advanced nodes, particularly its 3nm gate-all-around (GAA) process, which delayed mass production. The company has been investing heavily to improve its 2nm GAA technology, targeting production readiness by 2025. A $200 billion commitment from Broadcom, one of the largest custom chip designers, could provide the volume and revenue stability needed to accelerate yield improvements and attract other major customers. The deal also helps Broadcom reduce its reliance on TSMC for leading-edge manufacturing, adding supply chain resilience at a time when geopolitical tensions and natural disasters pose risks to concentrated production.
Advanced packaging: Integrating memory and logic
The collaboration extends to advanced packaging that integrates memory and logic more tightly, a technique increasingly central to squeezing more performance out of AI silicon. Advanced packaging, such as 2.5D and 3D chip stacking, allows different types of chips to be combined into a single package, reducing latency and power consumption while increasing bandwidth. Samsung has been expanding its advanced packaging capabilities, including its H-Cube (Hybrid Cube) and I-Cube technologies, to compete with TSMC’s CoWoS (Chip-on-Wafer-on-Substrate) and InFO (Integrated Fan-Out) platforms. By offering a combined memory, foundry, and packaging solution, Samsung can provide a one-stop shop for AI chip design and manufacturing, similar to TSMC’s integrated approach but with the added advantage of in-house HBM production.
Broadcom’s role in the AI chip ecosystem
Broadcom designs custom AI chips for hyperscalers including Google, Meta, and others that want silicon tailored to their specific workloads rather than buying off-the-shelf GPUs from Nvidia. These custom chips, often called accelerators or TPUs, are optimized for tasks like training large language models, inference, and recommendation systems. Broadcom has extended its AI chip partnership with Meta through 2029, and CEO Hock Tan has stated that surging AI revenue has made organic growth more attractive than acquisitions. By diversifying manufacturing to Samsung, Broadcom gains supply chain resilience beyond its existing reliance on TSMC. This move is particularly important as TSMC faces capacity constraints due to soaring demand from Nvidia, AMD, and other AI chip makers. Having a second source for leading-edge production could help Broadcom secure the output needed to meet its customers’ aggressive deployment timelines.
Samsung’s foundry credibility gap
For Samsung, the deal addresses a persistent credibility gap in its foundry business, where its global market share has fallen to roughly seven percent while TSMC dominates. The company has struggled with yields on its two-nanometer process, but a $200 billion customer commitment from one of the industry’s largest chip designers could change the calculus for investors who have questioned whether Samsung can compete at the leading edge. Samsung posted record quarterly operating profit of 89 trillion won ($67.5 billion) in Q2 2024, driven almost entirely by AI memory demand. However, its foundry division has been a drag on overall profitability. The Broadcom partnership could provide the volume needed to improve yields and reduce costs, making Samsung a more viable alternative to TSMC for other major customers such as Qualcomm, AMD, or Apple. Industry analysts note that Samsung must also navigate potential conflicts of interest, as it competes with Broadcom in some networking and wireless chip markets, but the MOU appears to focus on AI-specific products where overlap is limited.
Broader context: South Korea’s semiconductor push
The signing took place alongside a broader package of semiconductor deals worth roughly $950 billion between South Korean and American companies, announced during an AI summit hosted by South Korean President Lee Jae Myung. SK Hynix signed memory partnerships worth $750 billion with Nvidia and others, and Anthropic, an AI safety startup, signed supply agreements with both Samsung and SK Hynix. The summit followed Seoul’s $880 billion domestic investment plan unveiled in June 2024, which committed Samsung and SK Hynix to building four new chip fabrication plants in the country’s southwest region, known as the K-Semiconductor Belt. This massive investment aims to secure South Korea’s position as a global semiconductor powerhouse, especially in AI memory and advanced packaging. The deals also highlight the deepening integration between US tech firms and Korean manufacturers, as Washington seeks to onshore critical chip supply chains while balancing partnerships with allies.
The Samsung-Broadcom partnership is a testament to the accelerating demand for AI infrastructure. As hyperscalers like Google, Meta, and Microsoft rush to deploy large language models and generative AI services, the need for specialized silicon has skyrocketed. Custom AI chips offer better performance per watt and lower total cost of ownership compared to general-purpose GPUs, making them attractive for large-scale deployments. Broadcom’s ability to design these chips at scale, combined with Samsung’s capacity to produce them using cutting-edge memory, logic, and packaging technologies, creates a powerful vertically integrated supply chain that could challenge the dominance of Nvidia and TSMC. However, execution risks remain: Samsung must deliver on its 2nm technology promises, and Broadcom must prove that its custom chips can compete with Nvidia’s roadmap in terms of performance and ecosystem support.
The MOU also has implications for the broader semiconductor industry. If successful, it could encourage more chip designers to dual-source their manufacturing, reducing the near-monopoly of TSMC at leading nodes. This diversification is critical for national security, as seen in the US CHIPS Act and similar initiatives in Europe and Japan. For South Korea, the deal reinforces its role as a global hub for AI memory and manufacturing, potentially attracting more design and R&D investments from Silicon Valley. For Broadcom, it secures supply for the next five years at a time when AI chip demand is outpacing production capacity. The partnership’s true impact will become clearer as the first HBM4 and 2nm products roll out in 2025-2026, but the sheer scale of the commitment already signals a new era of collaboration in the AI chip industry.