A Coincidence That Is Not Really a Coincidence

On July 23, Nvidia agreed to prepay Amkor Technology $1.5 billion to expand advanced chip packaging and test capacity in the United States, mostly at Amkor's Arizona campus 1 3. Nineteen days later, Bloomberg reported Amkor was exploring a sale of a stake in its China packaging business — a unit also valued at roughly $1 billion to $1.5 billion 2. Amkor has hired an adviser to prepare a carve-out and may keep a minority position 2. The company has not framed the two figures as a trade-off. But the timing lays out, almost too neatly, where the world's largest outsourced chip packager is putting its capital: toward the customer with the deepest pockets in AI, and away from a market where US export controls have made that customer's chips increasingly hard to sell.

For investors tracking Asian supply chains, the more interesting reaction is not Amkor's. It is what China's own packaging industry has done in the same eight weeks.

China's Packaging Firms Are Not Waiting for a Buyer

Jiangsu Changjiang Electronics Technology, known as JCET and the world's third-largest outsourced semiconductor assembly and test (OSAT) provider, reported first-half 2026 revenue of 19.53 billion yuan, a first-half record, with net profit up 79.4 percent year over year 5. Management framed the results explicitly around advanced packaging, and the company is not sitting on the cash. In June it committed 7.8 billion yuan, about $1.15 billion, to a new packaging and test plant in Shanghai's Lin-gang Special Area, with production lines for high-density redistribution layers, ultra-fine bumping and chiplet-style multi-die packaging aimed squarely at AI servers, HPC systems and GPUs 6.

JCET is not acting alone. TrendForce counts four Shanghai- and Shenzhen-listed packaging leaders — JCET, Tongfu Microelectronics, Huatian Technology and Forehope Electronic — announcing more than 27 billion yuan, roughly $4 billion, of new advanced-packaging capacity in the first half of 2026 alone, all pointed at AI computing demand 7. That is a segment that barely registered as a growth story two years ago, now moving at a pace closer to the memory or foundry cycle. South China Morning Post reporting on the buildout ties it directly to a mix of AI-driven order books, stronger first-half profits across the group, and Beijing's push for packaging self-sufficiency as Washington tightens the rules on who can buy what 4.

The Bottleneck Moved From the Wafer to the Package

The reason capital is flooding into packaging rather than only wafer fabs is that packaging, not raw transistor supply, has become the binding constraint on AI hardware shipments. TSMC's CoWoS process — the technology that stitches logic dies to high-bandwidth memory inside Nvidia's GPUs — was still running a supply-demand gap TrendForce estimated at roughly 20 percent as of mid-2026, even as TSMC races to lift monthly capacity from about 75,000 to 80,000 wafers toward a 120,000 to 130,000 target by year-end 8. That gap is exactly the opening Chinese OSATs are chasing on behalf of the customers TSMC cannot or will not serve: Huawei, Cambricon and the rest of China's domestic AI-chip ecosystem, which has no access to CoWoS and needs its own advanced-packaging supply to turn finished wafers into shippable accelerators.

What It Means for Positioning

The read-through for Asian markets is a bifurcating packaging supply chain rather than a single global shortage. On one side, Nvidia is underwriting Western and Taiwan-aligned capacity directly — the Amkor deal funds new US lines built to Nvidia's own packaging roadmap, with Amkor's Arizona site serving as TSMC's key domestic packaging partner 1 3 — concentrating the most advanced, highest-margin work with US-aligned suppliers and their customers. On the other, Chinese OSATs are building a parallel stack sized for domestic accelerators rather than Nvidia's architecture, funded by their own AI-driven profit surge rather than a single anchor customer's prepayment.

That split matters for how investors read OSAT earnings this cycle. JCET's profit growth and capex are not simply riding the same AI wave as Amkor's Nvidia contract; they are being driven by a distinct, protected demand pool that a US-based supplier structurally cannot serve, which is why JCET, Tongfu and Huatian can keep expanding even as Amkor retreats from the same geography. Watch utilization and pricing at those new Chinese lines as they ramp toward 2027 for the clearest signal of whether Huawei and Cambricon's chip-output targets are real or aspirational. And watch whether Amkor actually completes a sale of its China stake — that would be the clearest confirmation yet that the industry expects this split to be structural rather than a temporary rationing problem that eases once TSMC's capacity catches up.