Nvidia and Amkor Sign Advanced Packaging Agreement Worth Up to $3 Billion
nashnova research
Nvidia and Amkor's advanced packaging commitments now total $3 billion after a new $1.3 billion deal to expand 2.5D/3D capacity in South Korea — a sign that chip packaging has gone from back-end afterthought to a bottleneck big enough for major customers to lock down with multi-year cash.
What does a $3 billion packaging deal actually buy?
Amkor announced a new $1.3 billion agreement with Nvidia, earmarked for expanding 2.5D/3D advanced packaging — the process of tightly connecting multiple chip dies inside a single package — capacity in South Korea.
Combined with earlier cooperation at Amkor's Arizona site, the two companies' multi-year commitments now total roughly $3 billion.
This means → Nvidia is not just placing orders; it is pre-locking capacity. In the AI chip supply chain, whoever can package can ship — packaging is the bottleneck.
The stock had fallen 20% — what does this bounce signal?
Following the announcement, Amkor shares rose about 10% in pre-market trading on October 21. Before that, the stock had dropped 20.4% in October through the 20th — far worse than the Philadelphia Semiconductor Index's roughly 10.6% decline over the same period.
In plain terms = the market had been questioning whether packaging demand was real; Nvidia's long-term commitment served as a "the demand is real" signal, and the stock partially repriced.
But the bounce (+10%) still falls short of the prior drawdown (-20%), suggesting investors want to see actual utilization and yield data before fully re-rating.
TSMC, Amkor, Nvidia — how does the three-way split work?
In June, Amkor signed a ten-year agreement with TSMC under which TSMC purchases advanced packaging and testing services from Amkor, using InFO (integrated fan-out) and CoWoS — a packaging method that links compute dies to memory through an interposer — at the Arizona site.
The logic: TSMC makes wafers, Amkor packages and tests them, and their shared customers (like Nvidia) get onshore, end-to-end U.S. manufacturing.
This reflects a broader trend: AI chip production is forming a "foundry + packager" lock-in model — major customers use long-term contracts to secure the entire chain, rather than buying spot.
Why is the Arizona project already expanding to Phase 2?
Customer commitments exceeded Phase 1 planned capacity, so Amkor announced Phase 2 in September. Total planned investment for the Arizona campus rose to $12 billion, with Phase 2 expected to complete by late 2029.
This means → demand is not "just enough" — it is overflowing. Customers are willing to sign for capacity that does not yet exist, signaling a strong multi-year expectation of AI packaging undersupply.
Why is advanced packaging getting more expensive — and more critical?
The core architecture of an AI accelerator connects compute dies to high-bandwidth memory (HBM) through a 2.5D interposer; 3D integration goes further, using through-silicon vias (TSVs) — vertical connections punched through silicon — to shorten data paths.
As cloud providers keep scaling AI accelerator deployments, both packaging complexity and per-chip packaging spend are rising in tandem.
In plain terms = Amkor's growth runway has two engines: AI chip shipment volumes are climbing, and the packaging cost per chip is climbing too — volume and unit price both moving up.
Big orders in hand — does profit automatically follow?
Nvidia's multi-year deal and the TSMC framework give Amkor solid customer and manufacturing-ecosystem backing.
But contracts lock in the revenue side; the profit side hinges on capacity utilization and yield ramp — new lines face a clear gap between first output and full production, and between initial yield and mature yield.
This means → the next milestone investors should watch is not "are there orders?" but "once the new capacity is running, can costs be controlled?"
市场有风险，内容仅供研究参考，不构成投资建议。