Why packaging became the bottleneck
The advanced packaging technology most widely associated with this capacity crunch is Chip-on-Wafer-on-Substrate, or CoWoS, though it is important to note that the official filings describe the work only as “advanced packaging” without naming CoWoS specifically. CoWoS is the most likely candidate given the context: it allows chipmakers to stitch together multiple processor dies and high-bandwidth memory stacks into a single package, delivering the massive data throughput that modern AI workloads require. Nvidia’s data-center GPUs, from the H100 through the Blackwell generation and the upcoming Rubin architecture, all depend on some form of advanced packaging to hit their performance targets. For most of the semiconductor industry’s history, packaging was a quiet back-end step, something that happened after the real engineering was done. That changed as AI training clusters grew from dozens of GPUs to tens of thousands. Each accelerator needs its own advanced package, and Nvidia’s production volumes have scaled so aggressively that TSMC’s in-house packaging lines became a binding constraint on the entire AI chip supply chain. Industry reporting from Reuters, Nikkei Asia, and Bloomberg has consistently identified Nvidia as the largest single source of advanced packaging demand, though neither TSMC nor Nvidia has publicly confirmed specific booking volumes. “We are seeing unprecedented demand for advanced packaging capacity,” TSMC CEO C.C. Wei said during the company’s January 2025 earnings call, adding that the company was working with partners to expand supply. Wei did not name Nvidia or CoWoS specifically, but the remark aligned with the timing of the Amkor partnership expansion.What the Amkor deal actually says
Two corporate filings from Amkor on BusinessWire lay out the confirmed facts. The October 2024 release announced the expanded TSMC partnership and framed it as a direct response to rising customer demand for high-volume advanced packaging. The October 2025 release confirmed the groundbreaking and the $7 billion investment figure. What the filings do not include is equally important. There is no breakdown of how much packaging volume is being transferred from TSMC to Amkor, no percentage split of work between the two companies, and no timeline for when the Arizona facility will begin qualified production. The filings reference customer demand in general terms but do not name Nvidia or any other specific buyer. The $7 billion figure represents Amkor’s total investment commitment for the campus, not a single capital expenditure or a payment from TSMC. How much has already been spent, and over what time horizon the rest will be deployed, remains undisclosed. The specific packaging technologies that will run at the Arizona campus are also unspecified. CoWoS is the most likely candidate given the context, but Amkor describes the site broadly as an advanced packaging and test facility. Whether it will handle only overflow from existing TSMC processes or eventually take on next-generation formats is an open question.The Arizona factor
The location is not incidental. TSMC is already building its own wafer fabrication complex in Arizona, known as Fab 21, with support from the CHIPS and Science Act. Placing an advanced packaging campus nearby creates a potential end-to-end production corridor on American soil: wafers fabricated at TSMC’s Arizona fabs could, in theory, move to Amkor’s adjacent campus for packaging and test without crossing an ocean. That geographic proximity matters for two reasons. First, it shortens the supply chain for U.S.-based customers and reduces exposure to the geopolitical risks that come with concentrating critical semiconductor capacity in Taiwan. Second, it gives major chip buyers like Nvidia, AMD, and the hyperscale cloud providers a domestic sourcing option for the most advanced packaging steps, something that did not exist before this partnership took shape. The deal also reshapes the competitive landscape for packaging. Intel has been investing in its own advanced packaging capabilities as part of its foundry services push, and Samsung has expanded its packaging offerings in South Korea. But the TSMC-Amkor arrangement is the first to pair the world’s leading-edge foundry with a top-tier OSAT provider on a single U.S. campus, creating a combination that neither Intel nor Samsung can replicate today.What the overflow deal reveals about AI chip supply chains
TSMC’s decision to externalize work it once guarded closely is a structural shift, not a temporary patch. Advanced packaging has moved from a back-end afterthought to a strategic chokepoint that can dictate how quickly the AI hardware buildout progresses. When the world’s most capable foundry concludes it cannot build advanced packaging capacity fast enough on its own, that tells the market something about the scale of demand that Nvidia and its peers are placing. For investors tracking this space, the key metrics to watch will come from Amkor’s quarterly earnings: capital expenditure disclosures, utilization rates once the facility begins operations, and any commentary on customer qualification timelines. TSMC’s own earnings calls will also be worth monitoring for language about advanced packaging capacity and partner contributions. For the broader industry, the takeaway is that the packaging supply chain is diversifying out of necessity. Arizona is becoming a second node for advanced packaging capacity outside Taiwan, and the companies that control that capacity will hold significant leverage over the pace of AI chip deployment for years to come. More from Morning Overview*This article was researched with the help of AI, with human editors creating the final content.
