Jake Sullivan on Navigating AI Uncertainty and Managing Competition with China

Episode Summary

Executive Summary: Jake Sullivan argues the U.S. should treat transformative AI as a serious near-term possibility while focusing policy on concrete risks and opportunities in security, economics, and society. He favors managed competition with China, selective export controls on high-end chips, greater government AI adoption, and deeper AI-risk dialogue, while rejecting regime change, grand bargains, and broad decoupling.

Main Topics: AI timelines and policy posture (Priority: 5/5): Sullivan says transformative AI could arrive within a couple of years, but he is not certain it will and prefers planning assumptions over certainty. He emphasizes concrete risk management over speculative debates about AGI/ASI or existential scenarios. U.S. government adoption of AI (Priority: 5/5): He argues the national security establishment is behind the frontier in integrating AI and needs leadership-driven adoption across agencies, with Congress still lagging in urgency. Export controls, compute, and Middle East deals (Priority: 5/5): Sullivan supports exporting chips only under strict security conditions, wants frontier training runs in the U.S., and worries the Trump administration’s Gulf deals and H-20 decisions conflict with the stated goal of staying ahead of China. China as a strategic competitor (Priority: 5/5): He frames China as the only country capable of matching and surpassing the U.S. across economics, tech, military, and diplomacy, citing overcapacity, coercive economic power, cyber prepositioning, and military buildup. Managed competition over decoupling or grand bargains (Priority: 4/5): Sullivan rejects regime change and broad decoupling, favoring a stable equilibrium based on intense competition, diplomacy, AI-risk dialogue, and guardrails to prevent conflict. AI’s domestic impacts on work and society (Priority: 4/5): He stresses job displacement, wealth concentration, misinformation, alienation, and the need for social insurance, while also noting major opportunities in productivity, health, and personalized services. Military and geopolitical lessons from AI (Priority: 4/5): He sees Ukraine and other conflicts as early evidence of AI-enabled warfare, logistics, intelligence, and autonomous systems, and worries the PLA may adopt these capabilities faster than the Pentagon.

Key Arguments: Transformative AI is a distinct near-term possibility and should be planned for, but policy should not assume arrival by a fixed date. The U.S. is ahead at the frontier but behind in real-world adoption of AI across national security agencies. Congress is still not operating with enough urgency on AI; it typically reacts to real-world shocks rather than abstractions or demos. Export controls on high-end chips are justified as a national security measure because advanced compute is a strategic input, especially given China’s civil-military fusion. The U.S. should not become dependent on foreign regions, including the Gulf, for compute; frontier training should remain largely in the United States. China poses unique risks through industrial overcapacity, coercive economic leverage, cyber prepositioning in critical infrastructure, and a massive military buildup. The goal should be managed competition with China, not regime change, Cold War escalation, or a grand bargain. AI policy should focus on concrete risks and opportunities in security, economics, and society rather than abstract existential debates alone. The government should better understand specific downside scenarios, because decision-makers need clear threat models and concrete response plans. A positive vision for AI’s future is missing from public debate, and policymakers should articulate one alongside risk mitigation.

Data Points: Near-term AI possibility: "the next couple of years" - Sullivan says transformative powerful AI could plausibly arrive in that window and should be taken seriously. AGI/ASI timeline examples: "27, 28, or even 2030" - He says he does not believe AI is inevitable by those dates. National security memorandum: Issued toward the end of his tenure - He describes a lengthy memorandum directing U.S. national security agencies to accelerate AI adoption. Congressional urgency gap: Qualitative gap; no number given - He says Congress remains behind where it needs to be on AI urgency. Gulf compute buildout: "five-gigawatt data center" - Raised as an example of potentially dangerous large-scale frontier infrastructure outside U.S. control. China Shock reference: "China Shock 2.0" - He cites David Autor and co-authors warning a second, potentially worse wave of trade and industrial disruption. AI action plan timing: Discussed "a week after" release - The interview references the Trump AI action plan as a recent development. China military buildup: "the world's largest peacetime military buildup" - Sullivan uses this as evidence of China’s strategic ambition and threat posture. Ukraine conflict lesson: "wires to the drones" and "PlayStation controllers" - He describes the current stage of AI-enabled warfare as still partly human-operated but increasingly autonomous.

Pivotal Quotes: ""Boring may be best in many scenarios."" — Grant/Ghosted host commentary: A framing remark about Sullivan’s pragmatic, restrained approach to AI and China policy. ""We have to take seriously the possibility that of AI 2027."" — Jake Sullivan: He says he has read and discussed short-timeline AI scenarios and treats them as credible possibilities. ""compete like hell, but intensively manage that competition so it didn't tip over into conflict"" — Jake Sullivan: His core description of the desired U.S.-China equilibrium.

Implications: Listeners should expect a pragmatic U.S. AI strategy: faster government adoption, targeted controls on frontier compute, and managed competition with China. The biggest open questions are whether policy can keep up with rapid AI progress and whether leaders can offer a compelling positive vision, not just risk reduction.