Samsung Electronics said it has signed a five-year memorandum of understanding with Broadcom worth more than $200 billion, covering sub-2nm foundry work, HBM4 and HBM4E supply, and advanced packaging through 2030.
The MOU was signed on July 24 during an AI summit in San Francisco and was announced by Samsung on July 25. According to the report, Samsung will supply HBM4 and HBM4E high-bandwidth memory for Broadcom’s next-generation AI accelerators. On the manufacturing side, Samsung will produce Broadcom-designed ASICs using a process below 2 nanometers.
Memory, foundry and packaging are all included
The agreement also reaches into advanced packaging, including 2.3D and 2.5D integration. The report says the order did not emerge in isolation, but came as South Korea has been pushing semiconductor and AI infrastructure cooperation more aggressively.
South Korean President Lee Jae-myung was in Silicon Valley at the time to host an AI summit. During that trip, he met Nvidia CEO Jensen Huang, OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, and Broadcom CEO Hock Tan. Samsung Chairman Lee Jae-yong, SK Group Chairman Chey Tae-won, Hyundai Motor Group Chairman Chung Euisun, and Naver founder Lee Hae-jin were also present. The event concluded with the release of the San Francisco AI Declaration.
Partnerships involving Nvidia and Naver, as well as Nvidia and SK hynix, were also announced during the same trip.
The order is far larger than Samsung’s annual foundry revenue
The article says Samsung and SK hynix have spent the past six months pushing hard for AI chip orders as Taiwan Semiconductor Manufacturing Co. continues to widen its lead in advanced foundry services.
According to the figures cited in the report, TSMC held nearly 70% of the global pure-play foundry market in 2025, with revenue of $122.5 billion. Samsung’s share stood at 7.2%, with revenue of $12.6 billion. On that basis, Broadcom’s $200 billion order over five years is worth more than 15 times Samsung’s full-year 2025 foundry revenue.
The report argues that this does not necessarily mean Samsung has suddenly overtaken TSMC on technology. A more likely explanation, it says, is that Broadcom does not want to depend on TSMC alone. Broadcom has expanded by designing AI ASICs for cloud giants such as Google and Meta. As volumes grow, relying on a single foundry leaves Broadcom exposed to pricing decisions and production prioritization set by one supplier.
Adding Samsung as a second source fits the industry’s usual risk-diversification logic, though the size of this agreement makes the move stand out.
South Korea’s government also features in the backdrop
The report says Lee’s Silicon Valley trip put diplomatic weight behind semiconductors and AI infrastructure. The broader South Korea-U.S. cooperation package tied to that visit was described as being worth about $950 billion, and it included an SK hynix-Nvidia investment project.
At the summit, Anthropic CEO Dario Amodei said the company had signed supply agreements with Samsung Electronics and SK hynix. Anthropic, which raised capital in May at a post-money valuation of $965 billion according to the report, named Samsung, SK hynix and Micron as key future suppliers of memory and chips.
In the report’s framing, geopolitics, capacity anxiety and capital all came together behind the Broadcom-Samsung deal.
Yield remains the critical test
The article also draws a line between the headline value of the MOU and what Samsung may actually ship. It says Samsung’s 2nm mass-production yield has reportedly been around 50% to 60%, still close to the threshold for stable volume manufacturing. By contrast, TSMC’s yield for the same generation is said to have moved above 70%.
That leaves yield improvement as the main factor that will determine whether Samsung can convert the framework into real output over the next few years. The report says the agreement looks more like an entry ticket than a guarantee for Samsung and SK hynix.
If Samsung fails to close the yield gap, Broadcom could shift order weighting back toward TSMC. The article adds that South Korea’s goal of doubling memory capacity within five years and building what it calls world-class manufacturing capability will still have to be backed by wafer-by-wafer yield performance, not summit optics.