Andrew Ng: US Policy Drives Sovereign AI
Andrew Ng warns that U.S. policies like sanctions, export controls, and restrictive immigration are accelerating a global push for sovereign AI capabilities. He argues these measures inadvertently boost open-source competition as other nations seek to reduce their reliance on American technology. This geopolitical dynamic is reshaping the competitive landscape for both closed and open AI ecosystems.
U.S. export controls on AI-related chips and semiconductor tools to China were first significantly expanded in October 2022. These policies aimed to restrict China's access to high-end GPUs like NVIDIA's A100/H100, which are essential for training large-scale AI models. Despite this, a policy shift in January 2026 began allowing the export of some advanced chips to China on a case-by-case basis, accompanied by a 25% tariff. This policy uncertainty has galvanized numerous countries to build their own "sovereign AI" capabilities, defined as a nation's ability to produce AI using its own infrastructure, data, and workforce. Nations like India, the UAE, Canada, and Japan are investing heavily in domestic compute infrastructure. India, for instance, aims to have over 38,000 GPUs in 2025 for domestic startups and researchers and is also working on developing its own GPUs by 2030. The global open-source AI model market is projected to grow from $19.05 billion in 2025 to $50.03 billion by 2030. Andrew Ng argues that participating in the open-source ecosystem is the most cost-effective route for nations to achieve AI autonomy without having to build everything from scratch. This trend is mirrored in enterprise adoption, where over 60% of AI projects now integrate open-source models for greater flexibility and to avoid costly proprietary software. This push for sovereign and open models directly impacts API strategy, especially with the rise of agentic AI. Agentic architectures transform LLMs into autonomous agents that can reason, plan, and execute multi-step tasks with minimal human input. This necessitates a shift from traditional APIs with predefined endpoints to more flexible, context-aware interfaces that can support dynamic, goal-oriented agentic workflows. As enterprises deploy these more autonomous systems, AI governance becomes a critical enabler rather than a constraint. Frameworks like the NIST AI Risk Management Framework and ISO 42001 provide structured approaches for managing risk, ensuring legal compliance, and maintaining stakeholder trust. Effective governance spans the entire AI lifecycle, from data collection and model validation to ongoing monitoring, which is crucial for operating in regulated industries. However, enterprise AI adoption faces significant hurdles beyond technology. While 72% of enterprises have adopted at least one AI capability, major challenges include inaccessible or low-quality data, a lack of technical expertise, and misalignment between business and technical goals. Case studies reveal that successful scaling requires embedding AI into core workflows with a clear connection to business value, not just celebrating pilot projects. U.S. immigration policy adds another layer of complexity, with current laws often favoring large companies and creating an uncertain climate for foreign AI talent. The U.S. government itself is expanding its use of AI in immigration adjudications, with systems like StateChat and ImmigrationOS being used to review petitions and detect anomalies across large datasets. This algorithmic review process increases the need for policy-aligned, rigorously documented visa applications.
