Samsung and Broadcom sign $200 billion deal on memory and chip manufacturing
Samsung and Broadcom have signed a memorandum of understanding to expand their strategic collaboration in memory and semiconductor manufacturing, with the goal of sustaining the next generation of artificial intelligence infrastructure. The announcement came during the AI Summit held in San Francisco, with the presence of Jinman Han, head of Samsung’s foundry business, and Hock Tan, CEO of Broadcom, along with representatives of the South Korean government.
The agreement does not come from nowhere. Samsung had already been supplying high-bandwidth memory to Broadcom for months, in parallel with the initial 12-layer HBM3E order that the Korean company closed with NVIDIA. This contract served to validate Samsung’s manufacturing quality before one of the most demanding clients in the sector. Now, Broadcom is establishing itself as a second major advanced memory customer, reducing Samsung’s dependence on a single buyer.
The two companies estimate that the joint collaboration in memory and manufacturing will exceed the 200 billion dollars in the next five years, until 2030. It is a figure that places this agreement among the most ambitious in the semiconductor sector related to artificial intelligence, and reflects the extent to which the demand for specialized chips is reorganizing alliances between manufacturers.
HBM memory for Broadcom AI accelerators
In the memory section, Samsung will supply HBM solutions for Broadcom’s future artificial intelligence accelerators. This type of memory stacks several layers of chips vertically to offer more bandwidth than conventional memory, an essential requirement in systems that train large models. The demand for HBM has grown so fast that manufacturing it with stable yields has become one of the main sector bottlenecks in the midst of a brutal component crisis.
Broadcom does not manufacture its own artificial intelligence chips to sell directly, but rather designs custom accelerators for large customers. For example, among them is OpenAI, which recently commissioned Broadcom to manufacture its own AI chips to deploy ten gigawatts of computing power starting in 2026. Ensuring sufficient memory for projects of this caliber is precisely one of the reasons behind the new agreement with Samsung.
2 nanometer manufacturing and advanced packaging
The second block of the agreement affects the foundry business, that is, the manufacturing of chips to order. Samsung will produce Broadcom products, including broadband wireless communications solutions, using its 2 nanometer process or less. Reducing the size of the transistor allows more computing capacity to be placed in the same space and lower energy consumption, something especially relevant in network chips and connectivity for data centers.
The collaboration will also include advanced packaging techniques on that same 2-nanometer process, such as 2.3D and 2.5D integrationwhich allow different chiplets to be combined in the same package. It is the same philosophy that NVIDIA or AMD use in their most powerful accelerators, and confirms that the performance of a chip depends less and less on the silicon itself and more on how it is assembled together with other components.
For Samsung, closing a client the size of Broadcom is especially relevant in its foundry business, where it competes against the historical dominance of TSMC. Ensuring volume at the 2 nanometer node helps make investment in new production capacity profitable, at a time when manufacturers like Samsung need anchor customers to sustain their most advanced factories for years.
On the other hand, the presence of representatives of the South Korean government in the announcement is not coincidental. South Korea considers the semiconductor industry a strategic priority and agreements of this caliber reinforce Samsung’s position against competitors such as TSMC or SK Hynix in full expansion of demand linked to the artificial intelligence.
