Commentary: US delays DeepSeek blacklist, exposing AI export-control dilemma

Washington’s pause in blacklisting DeepSeek and CXMT reveals a structural flaw in its AI export control regime: an inability to precisely target without triggering systemic supply chain fractures. Technically, this delays disruption in HBM and advanced packaging ecosystems but accelerates China’s pivot toward RISC-V and in-memory computing architectures. Compliance burdens have become nonlinear—even without formal sanctions, implicit scrutiny on U.S.-origin EDA tools and equipment forces Chinese foundries to rebuild entire validation workflows. Competitors like TSMC and Samsung may tighten IP licensing, while NVIDIA could deploy downgraded chips to retain market access. Over the next 18 months, U.S. policy will oscillate between surgical decoupling and overreach, inadvertently catalyzing China’s self-reliant AI inference stack on mature nodes (28nm+), locking both sides into a negative feedback loop of restriction and counter-innovation.