Samsung and Broadcom sign $200B five-year AI semiconductor pact

At the San Francisco AI Summit, Samsung and Broadcom signed a memorandum of understanding to construct an integrated AI semiconductor supply chain valued at $200 billion through 2030. Samsung's contributions span the stack: HBM4 and HBM4E high-bandwidth memory, sub-2nm foundry processes, and advanced 2.3D/2.5D packaging — the full set of technologies that gate cutting-edge AI accelerators.

The deal is another entry in the week's megadeal theme, but on the supply side rather than the compute-buyer side. It positions Samsung as a comprehensive counterweight to TSMC-plus-SK-Hynix in the AI hardware chain, and gives Broadcom (a major custom-silicon and networking player) a committed memory-and-foundry partner as demand for bespoke AI chips surges.

The financial backdrop validates the timing: Samsung reported Q2 2026 operating profit surging roughly 19x year-over-year, driven by runaway memory pricing and AI demand, on revenue of about 171.5 trillion won (~$118.56 billion). Samsung stock rose on the results, and the company separately reiterated a goal of expanding Galaxy AI to 800 million devices by end-2026 via a multi-provider strategy (in-house Gauss/Bixby plus Google's Gemini).

Caveats: an MOU is a framework, not a binding contract, and $200B 'through 2030' aggregates years of projected spend. HBM4/HBM4E and sub-2nm yields remain execution risks where Samsung has historically trailed TSMC. What to watch is whether the partnership translates into actual Broadcom design wins and whether Samsung's foundry can hit sub-2nm volume — the difference between a headline number and a genuine shift in the AI supply chain's balance of power.