Samsung Electronics signed a $200 billion supply agreement with Broadcom running through 2030, committing to deliver 2-nanometer process chips, high-bandwidth memory, and advanced packaging services for AI infrastructure buildout. The deal marks Samsung's largest public foundry commitment and its first named Tier-1 design win at the leading edge since losing Apple's A-series business in 2022.
Broadcom will take delivery of 2nm logic alongside fourth- and fifth-generation HBM stacks, with Samsung providing integrated packaging at its newly expanded Pyeongtaek and Hwaseong facilities. The contract backstops Samsung's 625 trillion won regional semiconductor investment plan announced the same week, targeting Yeongnam, Honam, and Chungcheong provinces for fabrication expansion. Samsung begins 2nm pilot production in the second half of 2025, with volume ramp scheduled for early 2026—Broadcom's deal provides the demand anchor that justifies full-line utilization before hyperscaler commitments typically arrive.
The timing matters because TSMC controls roughly 62% of the advanced logic foundry market, and its Arizona 2nm fab won't reach high-volume manufacturing until late 2027. Broadcom designs custom ASICs for Google, Meta, and ByteDance's AI training infrastructure, meaning this Samsung deal effectively diversifies the entire hyperscale supply chain at the most critical process node. Samsung's HBM3E and HBM4 memory will ship vertically integrated with the logic die, a packaging capability TSMC must still coordinate with SK hynix or Micron. For allocators, the structure implies Broadcom sees yield risk at TSMC Arizona or simply refuses to depend on a single supplier for chips that cost $15,000 per unit at volume.
Samsung's foundry division posted a $3.1 billion operating loss in 2024, and the Broadcom contract won't erase that immediately—2nm gross margins typically run negative in the first year as yields climb from 60% to the mid-80s required for profitability. But the six-year horizon gives Samsung the cash-flow visibility to justify $90 billion in incremental CapEx across the three named regions, and it forces Intel Foundry and Global Foundries to explain why their 18A and 12nm-class offerings remain relevant when the hyperscalers are locking 2nm at scale. The deal also binds Broadcom to Samsung's gate-all-around transistor architecture, which diverges meaningfully from TSMC's nanosheet implementation—any future design wins will carry switching costs measured in quarters, not weeks.
Operators should track Samsung's 2nm yield reports in Q4 2025 and whether Broadcom's order timing shifts forward, which would signal tighter-than-expected TSMC allocation or stronger-than-forecast AI ASIC demand. Foundry booking lead times for 2nm currently run nine months, and any Broadcom pull-in would compress available capacity for other customers through 2027. The 625 trillion won regional build-out includes a new Gwangju fab, with groundbreaking expected in Q1 2026—watch for co-investment announcements from Korea's national pension or sovereign development banks, which would confirm the government views this as strategic infrastructure rather than commercial risk.
Broadcom now holds the contract that makes Samsung's 2nm economics work, and Samsung holds the only high-volume alternative to TSMC for chips that don't exist yet but already have $200 billion in committed demand.