Table of Contents
Samsung offers Broadcom foundry, HBM, and packaging under one roof
In sum – what we know:
- A framework, not a contract – The MOU covers five years of cooperation the two companies expect to exceed $200 billion, with pricing and delivery negotiated project-by-project.
- One-stop AI silicon – Samsung will supply HBM4 and HBM4E, fabricate Broadcom’s logic on 2nm and below at Pyeongtaek, and handle 2.3D and 2.5D packaging.
- A geopolitical package – The deal sits inside a broader South Korea-US semiconductor and AI effort reportedly nearing $950 billion, aimed at allied supply chains.
Samsung Electronics and Broadcom have signed a five-year strategic memorandum of understanding that the two companies expect to exceed $200 billion in aggregate cooperation through 2030. The pact was signed at an AI Summit in San Francisco by Samsung Foundry President Han Jin-man and Broadcom CEO Hock Tan, with senior government officials from both South Korea and the United States in attendance — a signal that this is as much a diplomatic milestone as a commercial one.
This is a framework for long-term collaboration, not a single binding purchase contract. Detailed contracts, pricing, and delivery schedules will be negotiated project-by-project under the MOU, and there are no disclosed annual purchase targets.
Turnkey AI chip model
The core of Samsung’s pitch is a one-stop shop for Broadcom’s AI silicon, bundling design collaboration, foundry manufacturing, memory supply, and advanced packaging into a single integrated flow. Each piece matters on its own, but the integration is the actual selling point.
On the memory side, Samsung will supply its sixth-generation HBM4 and seventh-generation HBM4E to Broadcom, optimized for the extreme data throughput that AI training and inference workloads demand. On the foundry side, Broadcom’s advanced communications and AI logic chips will be fabricated on Samsung’s 2-nanometer and sub-2-nanometer process nodes, with production running at the company’s Pyeongtaek campus in South Korea. And on the packaging side, Samsung will use 2.3D and 2.5D techniques to tightly integrate logic chips and HBM into high-performance modules — placing compute and memory closer together to boost performance and cut power consumption.
The intended payoff is faster development timelines for Broadcom’s AI products and better power efficiency across the whole stack, since memory, logic, and packaging can be optimized together rather than in isolation. That’s the theory, at least. Samsung still has to prove it can deliver competitive yields at 2nm and below.
The comparison to TSMC is the obvious one. TSMC dominates leading-edge logic manufacturing, but it relies heavily on outside vendors for memory and, to a lesser extent, packaging capacity. Samsung is the only player that can fabricate the wafers, supply the HBM, and handle the packaging under one roof. Whether that vertical integration outweighs TSMC’s process lead is the open question — but for a customer like Broadcom, which needs all three pieces at scale, it’s a genuinely differentiated offer.
Geopolitics
For Samsung, the deal secures utilization of its most advanced nodes, underpinning multi-year capital expenditures at Pyeongtaek and providing revenue visibility in higher-margin AI and communications chips. A marquee customer committing to 2nm-and-below capacity is exactly what Samsung’s foundry business has needed. For Broadcom, the pact provides secured capacity at a leading-edge foundry and a dedicated memory partner — supply resilience that’s hard to overstate in a market where HBM and advanced packaging have become genuine chokepoints. It also diversifies Broadcom’s manufacturing footprint beyond any single foundry partner, and positions the company as a front-line competitor in AI accelerators and infrastructure rather than primarily a networking hardware vendor.
The broader context is political as much as commercial. The MOU forms part of a larger South Korea-US semiconductor and AI infrastructure package reportedly nearing $950 billion, involving other players including Nvidia and SK Hynix. It aligns neatly with efforts by both Washington and Seoul to strengthen secure AI chip supply chains among allied nations and reduce strategic reliance on semiconductor production in China.
The competitive response will be worth watching. TSMC, SK Hynix, and Micron now face a rival that has locked in a long-term, integrated relationship with one of the biggest custom silicon designers in the AI market. Expect them to pursue similar multi-year agreements rather than cede ground in HBM and advanced packaging. How much of the $200 billion actually materializes will depend on AI demand and Samsung’s execution over the next five years.
