Broadcom bets $200B on Samsung silicon through 2030

Broadcom just handed Samsung the biggest vote of confidence its foundry business has ever received. The two signed an MOU at an AI summit in San Francisco on 24 July covering HBM memory, 2nm-and-below foundry work, and advanced packaging (2.3D and 2.5D stacking) — reported at north of $200 billion through 2030.

The scope is what matters more than the number. Samsung supplies HBM for Broadcom's future AI accelerators, fabs Broadcom logic — including wireless broadband parts — on sub-2nm nodes, and packages memory tight against logic. That's a turnkey stack: memory, logic and packaging under one roof, which is precisely the pitch TSMC can't fully make on its own. Samsung Foundry has spent years short on marquee external customers; Broadcom is the world's top custom AI-chip designer, and its ASIC business is the main non-Nvidia path to hyperscaler silicon.

Caveats worth keeping: an MOU is intent, not a purchase order, and the headline figure traces to reports and Korean government framing rather than a hard contract disclosure. It also lands mid-memory crunch, with wafer capacity shifting to HBM and DRAM prices climbing.

Why it matters: if this converts, Samsung stops being the fallback fab and becomes a genuine second source for frontier AI silicon.