Executive Summary
Discussions of semiconductor competition often reduce U.S.-China technology rivalry to a race over process nodes: whoever advances first to the next leading-edge node is assumed to hold the advantage. This report argues that as Moore’s Law approaches its physical limits, the real axis of competition is shifting from transistor scaling to system-level integration. Advanced packaging sits at the center of this paradigm shift.
DSET argues that the U.S. export control regime has long followed a logic of “node-centrism,” focusing on advanced manufacturing equipment, especially extreme ultraviolet (EUV) lithography equipment, electronic design automation (EDA) tools, and performance thresholds for high-end artificial intelligence (AI) chips. This regime has indeed constrained China’s ability to access the most advanced semiconductor technologies. Yet it has also created a structural blind spot: when China cannot catch up at the frontier node on a single chip, it can still use advanced packaging to integrate multiple domestic dies that represent the best available within its process capabilities, achieving “good-enough” performance at the system level. Huawei’s Ascend series is a representative example of how system-level performance can be used to compensate for gaps at the process-node level.
This report dissects how China is turning advanced packaging into deployable AI compute by examining technology pathways, industrial policy, capability validation, and critical supply chain nodes. It also assesses the loopholes in the current export control regime. The report finds that China’s packaging breakout has made real progress, but remains constrained by engineering ceilings that cannot be quickly overcome. China is not simply seeking to compete head-to-head with the United States and its allies across every layer of technology. Instead, it is attempting to leverage its industrial foundations in mature-node manufacturing, outsourced semiconductor assembly and test (OSAT) capacity, substrates, printed circuit boards (PCBs), fan-out panel-level packaging, and domestic market demand to assemble “good-enough” compute capabilities sufficient to support domestic AI training and inference.
To compete effectively, the United States and its allies and partners must move beyond a narrow focus on frontier chips and frontier models. They must extend economic security strategy into the underlying industrial foundations that determine how AI capabilities are developed, deployed, and scaled. This report argues that export controls must move beyond a chip-by-chip framework and extend to the advanced packaging supply chain. To remain effective, controls must close gaps across four layers: materials, equipment and process technologies, finished products, and anti-circumvention enforcement.
1. Paradigm Shift: From Transistor Scaling to System-Level Integration
Advanced packaging improves integrated circuit performance, energy efficiency, and interconnect density by stacking and interconnecting multiple chiplets without requiring further advances in process nodes. The physical principle is straightforward: shorter chip-to-chip distances reduce signal latency, lower impedance loss, and decrease heat accumulation. This transformation is reshaping the semiconductor value chain and elevating back-end packaging into a core strategic technology in the competition over AI compute.
2. The Node-Centric Blind Spot: Regulatory Misalignment
U.S. export controls are centered on advanced nodes and chip performance thresholds. This logic has strategic necessity, but it has not adequately addressed China’s buildup in the back-end supply chain. It also overlooks the fact that advanced packaging and heterogeneous integration can generate meaningful system-level gains without access to frontier nodes. In other words, while export controls primarily target front-end manufacturing equipment and high-end AI chips, China is seeking compensatory pathways through back-end packaging interconnects and system-level integration.
3. Three Packaging Pathways and Asymmetric Breakout: “Tian Ji’s Horse Race”
Advanced packaging can be divided into three pathways according to the division of labor among firms. The first is foundry-led, such as TSMC’s CoWoS and SoIC, which offer the highest precision but also carry the highest costs. The second is foundry-OSAT collaboration, covering 2.5D and fan-out packaging. The third is OSAT-led, including FOCoS, FOPLP, and advanced FC-BGA, which offers lower costs and greater supply flexibility. China is currently concentrating on the third pathway and using it as a base for upward technological extension because it is cost-feasible, has lower exposure to existing controls, and can serve as a platform for accumulating engineering capabilities.
Table 1. Technology Characteristics of the Three Advanced Packaging Pathways
China’s logic of achieving more with less resembles the Warring States-era strategy of “Tian Ji’s horse race”: rather than matching the United States head-to-head in every technology domain, China is using asymmetric positioning to pursue system-wide advantage. This strategy rests on two pillars. First, China is using system integration as a substitute for process leadership, integrating the best available domestic dies at high density to narrow gaps at the system level rather than the wafer level. Second, it is leveraging downstream industrial depth, redirecting decades of accumulated capacity in packaging and testing, substrates, and PCBs to support more complex heterogeneous integration. This breakout pathway, however, still faces ceilings. Engineering bottlenecks such as interconnect precision, warpage control, and thermal management remain physical constraints as China attempts to move from mid- and lower-tier routes toward higher-density architectures.
4. Institutional Acceleration through a Whole-of-State System
China’s ability to turn advanced packaging from a technical possibility into deployable compute does not come only from isolated firm-level breakthroughs. It is built on institutional acceleration driven by national industrial policy, the Big Fund, policy finance, government procurement, and demand from national compute infrastructure. This system reduces investment uncertainty, absorbs yield-ramp and long-cycle technology iteration risks, and helps state-backed firms and national champions build a back-end ecosystem. In this sense, China’s advanced packaging pathway is not merely a technical choice, but an industrial mobilization model supported by policy, capital, and demand.
5. Capability Validation: Huawei’s Ascend Series
Huawei is one of the few Chinese firms spanning chip design, system integration, and data center deployment, making it a central case for testing this breakout pathway. From the Ascend 910C’s dual-die packaging architecture to the planned adoption of quad-die packaging in future products, Huawei’s trajectory shows that multi-chip integration has become a strategic direction for improving system-level compute under process-node constraints, even though Huawei’s products still remain behind Western frontier chips in important respects. Using total processing performance (TPP) and memory bandwidth as two analytical axes, the report examines the performance trajectory of future Ascend products and finds that their post-packaging system-level capabilities are challenging existing control thresholds.
Figure 1. Huawei Ascend Processors vs. Export Control Thresholds
Figure 2. AI Accelerator Performance Trajectories: Huawei Ascend vs. NVIDIA GPUs Across TPP and Memory Bandwidth (2025–2028)
6. Critical Supply Chain Nodes and Control Gaps: Where Vulnerabilities and Regulatory Gaps Overlap
This report disaggregates the advanced packaging supply chain layer by layer and finds that China’s most vulnerable points are precisely where current export controls have the weakest coverage.
- Layer One | Upstream Materials: AI and high-performance computing (HPC)-grade packaging materials such as ABF film, BT resin, and low-dielectric specialty glass fiber cloth are highly concentrated in Japan. China has not yet achieved high-end mass-production substitution, creating a priority control node where technical gaps and regulatory gaps overlap.
- Layer Two | Key Components: If upstream materials remain uncontrolled, controlling equipment alone is insufficient to prevent the accumulation of finished-product capabilities, especially in ABF substrates.
- Layer Three | Manufacturing Equipment: Although recent controls have expanded coverage of relevant equipment, certain back-end tools such as laser drilling, grinding, and automated inspection still face control gaps. At the same time, the pace of domestic substitution in China is narrowing the window for effective control.
- Layer Four | Advanced Packaging: Foundry-level platforms are subject to stricter controls, but the January 2025 OSAT due diligence rules mainly target transactions rather than capabilities. The equipment and materials used by OSATs still face control gaps, leaving room for circumvention.
- Layer Five | Final Systems: AI graphics processing units (GPUs) and high-bandwidth memory (HBM) are already controlled, but existing compute thresholds do not fully capture circumvention pathways based on chiplet integration and post-packaging system-level performance.
Table 2. Five-Tier Supply Chain Control Gap Overview
Policy Recommendations
This report argues that effective competition does not require abandoning export controls, but strategically reforming them to close loopholes, while complementing them with friend-shoring and allied division of labor. Based on its analysis of critical supply chain nodes, the report offers the following recommendations:
- Upstream Materials Controls: Controls should target highly concentrated supply nodes, including ABF film, BT resin, low-dielectric specialty glass fiber cloth, and advanced bonding materials, using technical thresholds rather than blanket bans. Export licensing should apply only to products that meet or exceed AI and HPC-grade advanced packaging specifications, distinguishing general commercial uses from high-end AI packaging applications. Because supply is highly concentrated in Japan, U.S.-Japan regulatory alignment should be a priority.
- Equipment and Process Technology Controls: Controls should establish clear quantitative thresholds for critical capabilities such as hybrid bonding, 2.5D and 3D heterogeneous integration, high-density substrate manufacturing, micro-bumps, and fine-pitch interconnects. These thresholds could include through-silicon via (TSV) density, redistribution layer (RDL) line width and spacing, micro-bump pitch, and substrate layer counts. This approach would be more precise and enforceable than relying on broad references to “advanced packaging technology,” while also improving coordination among allies.
- Finished-Product Controls: Control benchmarks should expand from single-chip specifications to post-packaging system-level performance. Controls should target aggregated products that exceed functional thresholds through chiplet architectures, 2.5D/3D packaging, or multi-chip module (MCM) integration. High-end FC-BGA substrates, AI and HPC modules, integrated systems, and all generations of HBM should be subject to licensing thresholds based on physical specifications and end use.
- Anti-Circumvention Architecture: A more complete anti-circumvention architecture should be built across four dimensions: supplier disclosure and due diligence obligations, anomalous order screening, third-country transshipment checks, and a high-risk transaction information-sharing framework centered on the United States, Japan, South Korea, and Taiwan. Third-country transshipment checks should prioritize back-end manufacturing destinations in Southeast Asia, such as Singapore, Malaysia, Vietnam, and Thailand.
In the long run, policy should move beyond fragmented unilateral industrial policy toward friend-shoring and a division of labor among democratic technology allies and partners. The comparative advantages of Taiwan, the United States, Japan, South Korea, and other democratic technology allies and partners should be integrated into a coordinated and resilient trusted supply chain. This report argues that promote and protect policies must proceed in parallel: industrial policy alone cannot offset China’s back-end industrial depth, and export controls remain indispensable as the United States and its allies and partners rebuild their own leverage.
