Samsung and Broadcom sign $200 billion memory and foundry deal for AI chips

Samsung Electronics and Broadcom announced on July 24, 2026 that they signed a memorandum of understanding to expand their strategic collaboration across memory and foundry technologies, in a deal the companies estimate at more than $200 billion over the next five years through 2030. The MOU was signed at the AI Summit held at The Midway in San Francisco, with Jinman Han, President and Head of Samsung Electronics' Foundry Business, and Broadcom CEO Hock Tan among the executives present, alongside representatives of the Korean government.

What the deal actually covers

The agreement spans two distinct parts of Samsung's semiconductor business. On memory, Samsung will supply High Bandwidth Memory — including HBM4 and the next-generation HBM4E — to support Broadcom's AI accelerators. HBM is the specialized, stacked memory architecture that sits directly alongside AI chips to feed them data fast enough to keep pace with modern accelerator compute, and it has become one of the most supply-constrained components in the AI hardware stack over the past two years.

On foundry, Broadcom will manufacture chips — including its wireless broadband communications products — using Samsung's 2-nanometer and below process technologies. The collaboration extends into advanced packaging as well, covering 2.3D and 2.5D integration built on Samsung's 2nm process, techniques used to combine multiple chip dies into a single package for better performance and power efficiency.

Why this matters for the AI supply chain

The deal arrives amid what Samsung's own executives describe as a severe memory shortage driven by AI infrastructure buildout. Every major AI accelerator maker — Nvidia, AMD, and now increasingly Broadcom's custom silicon customers — needs guaranteed HBM supply to ship products on schedule, and memory makers have struggled to expand capacity fast enough to match demand. Locking in a five-year, $200 billion framework gives both companies planning certainty: Samsung secures a committed customer for HBM4/4E output as it scales production, while Broadcom secures the memory and manufacturing capacity it needs to keep building custom AI accelerators for its hyperscaler customers.

Broadcom has increasingly positioned itself as the go-to partner for companies designing their own AI chips rather than buying off-the-shelf GPUs — a business that depends entirely on being able to source leading-edge memory and manufacturing at scale. This agreement gives Broadcom a second major foundry and memory partner alongside its existing relationships, reducing its dependence on any single supplier at a moment when AI chip demand shows no sign of slowing.

Samsung's broader positioning

"AI is driving unprecedented demand for tightly integrated semiconductor technologies spanning memory, logic and advanced packaging," said Young Hyun Jun, Vice Chairman and CEO of Samsung's Device Solutions Division, in the companies' joint announcement. The framing reflects Samsung's pitch to the industry: unlike pure-play foundries that only manufacture chips, Samsung can offer memory, logic, foundry, and packaging together, positioning it as a one-stop option for AI chip designers who need all four capabilities coordinated tightly.

For Samsung specifically, the deal also matters competitively. The company has trailed TSMC in leading-edge foundry market share for years, and a marquee customer commitment of this size at 2nm and below gives Samsung a concrete data point to point to as it tries to close that gap. As Samsung's official announcement notes, the collaboration is structured as a memorandum of understanding rather than a finalized binding contract, meaning specific volumes and pricing will still need to be worked out as the partnership progresses toward 2030.