The second session “AI Supply Chain Resilience Dialogue: Global Competition from Models and Security to Intelligent Robotics Applications” was moderated by Dr. Kai-Shen Huang, director of the Democratic Governance Program at DSET, and covered AI supply chain resilience. Speakers included Dean Ball, Senior Fellow, Foundation for American Innovation (FAI); Matt Cronin, Senior National Security Advisor, Andreessen Horowitz (a16z); Martijn Rasser, Vice President for Technology Leadership, Special Competitive Studies Project (SCSP); Jason Fiorillo, Chief Legal Officer and Secretary, Boston Dynamics; and Michael Robbins, President & CEO, Association for Uncrewed Vehicle Systems International (AUVSI). The discussion commenced in three parts: AI policy frameworks, AI diffusion, and “embodied AI.”
On AI policy, Mr. Dean Ball articulated the AI Action Plan, arguing that the U.S. is actively pursuing a vision of innovating in both open and closed weight models. He also discussed trends in the adoption of closed weight models by middle powers, contending that the United States needs to focus its efforts on exporting its AI stack globally—a goal that involves American allies adopting a common AI ecosystem.
Matt Cronin introduced a three-pronged strategy for the United States and its allies and partners to win the AI race against China: maintaining a sizable lead in frontier AI, facilitating model diffusion, and incentivizing human flourishing. He opined that all are interconnected and necessary components for American and allied victory at different layers of the AI stack.
On AI diffusion, Martjin Rasser contended that diffusion is the key race between the United States and China. He identified several dimensions of the diffusion race and alliance-building frameworks like Pax Silica. He also implored the United States to pursue regulatory predictability as a vehicle for effective coordination with European and East Asian allies and partners.
On embodied AI, Jason Fiorillo explained that countries will increasingly rely on intelligent robotics to ameliorate labor shortfalls caused by declining birthrates. He articulated Boston Dynamics’ mission to incorporate AI into robotics, sustain high customer satisfaction and demand, and maintain trust and security along the robotics supply chain.
Michael Robbins argued that the robotics industry highlights a key convergence between hardware and AI, noting that while the United States leads in the deployment of software, it does not in hardware. He conveyed the importance of alliances and partnerships in robotics, arguing that the United States is strongest when it works with likeminded countries—“America First” should not mean “America Only.” The Drone Dominance Program and cooperation with Taiwanese firms is a strong positive example. On Taiwan’s progress, Robbins expressed strong support for Taiwan’s Supplemental Defense Budget and spending more on asymmetric capabilities.
Highlights of the speakers’ presentations are as follows:
On AI policy, Dean Ball, a senior fellow at Foundation for American Innovation (FAI), discussed the Trump administration’s AI Action Plan, arguing that the U.S. is actively pursuing a vision of innovating in both open and closed weight models. While China leads in developing frontier open weight models, American open weight models, including Gemma and GPT-OSS, can catch up with Chinese models. Ball also mentioned the role of middle powers in the technological competition between great powers.
He stressed that the United States, who excels at closed weight models, should accelerate efforts in exporting the AI stack globally—a goal that involves American allies adopting a common AI ecosystem. He pointed to Taiwan as a shining example of how middle powers should respond to the mercurial nature of the Trump administration; middle powers should embrace frontier models to develop and iterate on technologies that can emerge as critical chokepoints and geopolitical leverage. Ball also touched upon AI governance and regulation. He proposed a “balanced approach” to regulation, which seeks to distribute liability evenly across the AI value chain lest it undermines development.
Ball cautioned against imposing too much precautionary regulation; instead, he recommended deepening societal resilience when managing concerns about the national security risks associated with AI. Finally, Ball offered some advice to the audience about engaging the U.S. on industrial issues during the second Trump administration. He advised them to not only communicate with Washington. Ball reminded the audience that there is a plethora of firms who are ready to collaborate with partners around the world.
Speaking on the same topic, Matt Cronin, a senior national security advisor at Andreessen Horowitz (a16z), opined that victory in AI competition comprises three prongs: winning at the frontier, establishing a global ecosystem, and the facilitation of human flourishing. He argued that all three conditions are mutually reinforcing—the United States and its allies and partners must win on all fronts of the competition. On the frontier, Cronin asserted that the United States and its allies and partners are creating the best models, hardware, and components. He noted that the United States will maintain the frontier lead because innovation yields a “cascading effect.” He also noted that China cares about competition over frontier AI, and Chinese AI labs are striving to close the gap with the United States.
Regarding diffusion, Cronin acknowledged China’s national integration and global deployment of models. Through massive investments in direct and indirect subsidies, China has excelled in diffusing its models globally, particularly in the global South. Cronin warned that such diffusion of Chinese models has created significant risks for democracies. Chinese models must be ideologically loyal to the Chinese Communist Party (CCP), they are vulnerable to data risks, and they are vectors for disseminating CCP propaganda. He noted the cognitive effects of the global diffusion of Chinese AI models and urged the United States and its allies and partners to spread the American AI stack broadly.
On human flourishing, Cronin identified how AI could be utilized to affect national and civilizational prosperity—a condition necessary for social cohesion and stability amidst technological change. He illuminated how AI tools, such as in the field of education, could enable a country’s citizenry to “unlock their true potential.” Observing the general lack of emphasis on this application of AI, he encouraged greater intentionality with deploying AI for this purpose.
Speaking on the state of AI competition, Martijn Rasser, the vice president for Technology Leadership at the Special Competitive Studies Project (SCSP), propounded that the AI race is not just a frontier race; it is a diffusion race. Rasser identified four arenas of the diffusion race: procurement, deployment, adoption, and scaling. He noted that while the United States is “very good at making technology,” it struggles to quickly deploy technology. On adoption, he observed that China excels at global diffusion because “good enough” models appeal to many countries.
To counter China’s lead in diffusion, the United States must engage its allies and partners with alliance frameworks, such as Pax Silica. Rasser identified Taiwan as central to American diffusion. Furthermore, he contended that the United States should encourage Europe’s development of emerging technologies, as this would create healthy competition and spur continued innovation. On a practical level, Rasser argued that the United States must create regulatory predictability, ensuring that the American AI becomes the global standard.
On embodied AI, Jason Fiorillo, Chief Legal Officer at Boston Dynamics, explained that countries will increasingly rely on intelligent robotics to ameliorate labor shortfalls caused by declining birthrates. The pace of physical AI development is extremely quick, with Fiorillo noting that the previous generation Atlas robot presented at CES last year was already obsolete. In industrial manufacturing, physical AI and humanoid robots will be critical in the final assembly step, which in the automobile industry, for example, has traditionally been hardest to automate.
Fiorillo agreed that AI could be an accelerant for physical applications, sharing Boston Dynamic’s ongoing work with Google DeepMind. He articulated Boston Dynamics’ mission to incorporate AI into robotics, sustain high customer satisfaction and demand, and maintain trust and security along the robotics supply chain. Critical to trust is “friend-shored” and “onshored” supply chains and critical components, providing allied alternatives to Chinese software, and a secure supply of key materials.
Michael Robbins, President and CEO of AUVSI, assessed that the United States is leading in software deployment but lagging in embodiment. Robbins argued that this problem is not unfamiliar: China’s industrial policies in other technologies like electric vehicles have also resulted in innovation flywheels, allowing a self-sustaining cycle creating technological advantage. Robotics, like other technologies, are inherently dual use, and Robbins highlighted recent research in the PLA’s efforts to weaponize robot wolves.
Allied partnerships are crucial, therefore, to ensure companies have the chance to compete and win on an even playing field, which was the impetus for AUVSI’s Partnership for Robotics Competitiveness. Robbins highlighted AUVSI’s work with ITRI and Taiwanese firms to build trusted drone partnerships, lending itself for Taiwan to play a greater role for an uncrewed arsenal of democracy at scale. To incentivize demand, Robbins argued that strong demand signals are important to accelerate manufacturing. Robbins noted the challenge posed by recent U.S. FCC action on foreign drone components, which hinders allied cooperation and should be more narrowly targeted. Finally, Robbins expressed strong support for Taiwan’s supplemental defense budget and spending in asymmetric capabilities.
