NVIDIA
2026-07-25
NVIDIA Prepay $1.5B to Amkor for US 2nm/3nm/HBM4 Advanced Packaging
Summary
Key Takeaways
NVIDIA signed a $1.5B advanced packaging deal with Amkor, prepaying to support capacity expansion in Tempe, Arizona, covering 2nm/3nm and HBM4 with 2.5D/3D packaging. This moves critical AI chip packaging from Taiwan to the US, mitigating tariff risks and securing supply chain.
Currently, NVIDIA relies on ASE (CoWoS-L) and TSMC CoWoS for HBM4 integration. Amkor, the #2 OSAT, brings FoWLP, InFO, 2.5D/3D capabilities, filling a US gap. Alongside Wistron Fort Worth (L6 integration), TSMC Arizona (wafer fab), Samsung Texas (wafer fab), and Micron Idaho (HBM), this completes a 'US AI manufacturing' chain.
Strategically, NVIDIA onshores packaging, reducing Taiwan dependency. The prepayment model signals deep financial binding in AI supply chain. Amkor's expansion will also serve AMD, Intel, Qualcomm, accelerating US AI capex localization from 5% to 15-20%. With demand from Vera Rubin (288GB HBM4) and MI455X (432GB HBM4), Amkor Arizona becomes the sole large-scale 2.5D/3D capacity in the US.
Why It Matters
NVIDIA's prepayment locks Amkor's US capacity, ostensibly for supply chain resilience, but actually defends against TSMC CoWoS and ASE CoWoS-L allocation uncertainties. By financially binding Amkor, NVIDIA secures priority access to 2.5D/3D packaging, encircling AMD and Intel in capacity competition.
For customers, NVIDIA's control over the only large-scale US advanced packaging facility locks them into NVIDIA's ecosystem (e.g., Vera Rubin, Blackwell Ultra), making switching costly. The higher US packaging cost (30-50% more) will be passed on, a cost trap downplayed.
Technically, Amkor's FoWLP/InFO maturity for HBM4 integration (8x12-Hi stacks) is unproven vs. TSMC's CoWoS-L, risking yield and thermal issues that could delay Vera Rubin.
PRO Decision
[Vendors] AMD should accelerate partnerships with ASE or JCET to secure diverse advanced packaging capacity, highlighting NVIDIA's lock-in risk and promoting open standards like UCIe. Intel can leverage its IFS and EMIB/Foveros packaging to offer alternative US-based packaging, directly challenging NVIDIA's supply chain narrative.
[Enterprises] CIOs should demand cost transparency and capacity guarantees from NVIDIA, while evaluating AMD MI455X and Intel 18A packaging independence. Include cross-chip portability clauses in contracts to avoid being locked into NVIDIA's packaging ecosystem. Implement multi-source packaging audits.
[Investors] Watch NVIDIA's prepayment impact on free cash flow and potential gross margin erosion (2-3 points) from higher US packaging costs. Amkor expansion benefits competitors, potentially diluting NVIDIA's packaging advantage. Long-term, onshoring raises costs, questioning NVIDIA's TCO leadership; reassess AI chip demand elasticity.
Get 3-5 key AI infrastructure signals weekly →
💬 Comments (0)