Samsung Lands a $200 Billion Chip Deal With Broadcom for the AI Buildout

Samsung Electronics has landed the kind of customer it has spent years trying to win. On Friday, July 25, at an AI summit inside The Midway in San Francisco, the South Korean giant and Broadcom signed a memorandum of understanding they value at more than $200 billion, a five-year pact running through 2030 that stretches across nearly every layer of the artificial-intelligence chip stack: high-bandwidth memory, leading-edge foundry manufacturing and advanced packaging.

If the numbers hold, it is one of the largest semiconductor partnerships ever disclosed. It is also, for now, a statement of intent rather than a binding contract, a distinction that matters more here than in most deals.

What the deal actually covers

The agreement splits into three pieces. On memory, Samsung will supply Broadcom with next-generation high-bandwidth memory, including HBM4 and HBM4E chips, the dense stacks of DRAM that feed data to AI accelerators fast enough to keep them from starving. On foundry, Samsung will manufacture Broadcom-designed products, including wireless broadband communications silicon, using its 2-nanometer-and-below process technology at its campus in Pyeongtaek, South Korea. And the collaboration extends into advanced packaging, including 2.3D and 2.5D integration built on that 2nm process, which lashes memory and logic together more tightly to squeeze out performance and power efficiency.

That end-to-end scope is the point. Broadcom is one of the world's leading designers of custom AI chips, the application-specific accelerators that hyperscalers such as Google, Meta and Apple commission instead of buying off-the-shelf GPUs from Nvidia. Broadcom designs the silicon; it needs partners to build and dress it.

"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. "By expanding our collaboration with Broadcom across these critical technologies, we look forward to delivering greater value to customers while advancing the AI infrastructure of the future."

Broadcom framed it as ecosystem insurance. "As AI infrastructure continues to scale, close collaboration across the semiconductor ecosystem becomes increasingly important," said Charlie Kawwas, president of Broadcom's Semiconductor Solutions Group. "By combining Samsung's memory and foundry expertise with Broadcom's AI and connectivity leadership, we aim to continue to deliver technologies that power the next generation of AI infrastructure."

The signing drew Broadcom CEO Hock Tan, Samsung Foundry president Jinman Han and representatives of the Korean government. It was one node in a much larger web: the summit, tied to South Korean President Lee Jae Myung's US visit, produced a roughly $950 billion package of Korea-US semiconductor commitments, with SK Group separately pursuing some $750 billion in long-term memory agreements anchored by SK Hynix's supply deal with Nvidia.

Why it matters: the foundry gap

For Samsung, the strategic prize is credibility. Its foundry business has been stuck at roughly a 7 percent share of the contract-manufacturing market while Taiwan's TSMC commands about 70 percent and more than 90 percent of true leading-edge production. Marquee customers have drifted away over yield concerns; Qualcomm reportedly moved 2nm work toward TSMC earlier this year. A public, $200 billion commitment from a designer of Broadcom's stature is exactly the vote of confidence Samsung's foundry unit has lacked.

The catch is execution. Samsung's 2nm yields have improved from roughly 20 percent in late 2025, but recent estimates put them in the mid-50 percent range, still short of the roughly 60 percent threshold the industry treats as the floor for stable mass production. TSMC is reportedly running 60 to 70 percent. An MOU commits Samsung to a target, not a track record; converting it into consistent high-volume 2nm output for AI chips is precisely the test Samsung has repeatedly struggled to pass. Analysts do not expect the foundry to turn a profit until around 2027 or 2028, as its Taylor, Texas plant ramps.

Why it matters: the HBM race

The memory half is the surer bet. Samsung shipped the world's first commercial HBM4 in February and has reportedly sold out its entire 2026 HBM4 capacity, with more than half of Pyeongtaek's 4nm capacity allocated to HBM logic base-die work. HBM4 demand is outrunning supply, and SK Hynix has led that market. Locking in Broadcom at HBM4 and HBM4E volumes gives Samsung guaranteed revenue and, just as important, justification for the tens of billions in capacity spending the company is pouring into chips this year.

For Broadcom, diversifying beyond TSMC buys supply-chain resilience as custom-ASIC demand explodes. Big Tech's appetite for in-house accelerators, tailored to specific workloads and cheaper at scale than merchant GPUs, is the engine behind this deal, and Broadcom sits at the center of it.

What to watch

- Whether the MOU converts into firm, binding orders, and on what timeline. Statements of intent at this scale have collapsed before. - Samsung's 2nm yield trajectory through 2027. The headline number is an ambition; the yield charts will decide whether it becomes a business. - The Taylor, Texas fab ramp and any move to produce Broadcom chips on US soil amid Washington's push to onshore advanced manufacturing. - TSMC's response, and whether other custom-silicon designers follow Broadcom in hedging their manufacturing bets. - First tangible milestones, tape-outs, qualified wafers, HBM4E shipments, that would signal the partnership is moving from press release to production.