The Research Institute for Democracy, Society, and Emerging Technology (DSET), together with Acropolis, co-hosted a Taiwan talk on December 16 featuring Jeffrey Ding, author of Technology and the Rise of Great Powers. Titled “In the U.S.–China AI Race, Does Taiwan Still Have a Future? Unpacking China’s AI Development—Its Strengths and Weaknesses,” the event was moderated by DSET Director Wen-Ling Tu and featured panelists including DSET CEO Jeremy Chih-Cheng Chang, CommonWealth Magazine Editor-at-Large Liang-Rong Chen, and Nikkei Asia Chief Technology Correspondent Cheng Ting-Fang. The discussion brought together perspectives from technology, media, policy, and international affairs to examine key variables in U.S.–China AI competition and Taiwan’s strategic options amid the restructuring of the global AI ecosystem and supply chains.
In her opening remarks, Director Tu noted that DSET’s engagement with Jeffrey Ding began with an interview last year. She shared that Ding is a professor at George Washington University and that his book, Technology and the Rise of Great Powers: How Diffusion Shapes Economic Competition, has sparked wide discussion in Washington, D.C. Over the past year, Ding has testified before the U.S. Congress multiple times and offered policy recommendations through platforms such as Foreign Affairs. Dr. Tu also recalled that DSET’s Deputy Director of the Democratic Governance Program You-Hao Lai invited Ding for an interview last year; after it was published, it generated broad discussion among Mandarin-language readers. When Acropolis released the Mandarin edition, Tu and Lai were also invited to write the foreword. Tu added that, in that foreword, she and Lai expressed admiration for Ding’s research approach in revisiting the three industrial revolutions, while also engaging critically with some of his policy recommendations, including those involving U.S. restrictions. She said the public forum was intended to deepen dialogue between the author and readers in Taiwan.
Jeffrey Ding: The U.S.–China AI Race Is a “Marathon,” Not a Sprint
In his talk, Ding explained that his book seeks to clarify how technological change translates into productivity advantages and, in turn, reshapes political and military power. Drawing lessons from three general-purpose technology waves—steam power, electricity, and computing—he presented two competing frameworks. The first is the “leading sector” approach commonly used in existing research, which treats competition as a sprint focused on dominating a handful of high-growth industries, controlling frontier innovation, and capturing monopoly rents. The second is his preferred “general-purpose technologies (GPT) diffusion” framework, which treats competition as a marathon: outcomes depend on whether a society can diffuse a GPT widely across sectors and sustain long-term adoption, organizational adjustment, and complementary innovation—thereby accumulating economy-wide productivity growth.
Ding further argued that if AI functions as a GPT, its full impact will take time to materialize. Assessments of national competition should therefore not overemphasize short-term breakthroughs, but instead focus on diffusion capacity and the institutional conditions that enable diffusion. He suggested that AI’s deeper effects on economic structure are more likely to become visible after 2030.
Turning to policy implications, Ding said that recent U.S. measures—such as export controls on advanced chips and restrictions on international student visas (including for Chinese students)—reflect a “Fortress America” approach centered on preventing technology leakage. From a GPT diffusion perspective, however, he argued that national economic strength cannot be determined solely by blocking innovation flows. Instead, he advocated a “run faster” strategy that prioritizes domestic diffusion and capability-building, including expanding AI deployment and long-term operations across diverse application settings, investing more heavily in the STEM workforce, and strengthening multiple education pathways—such as community colleges—to systematically build practical skills and implementation capacity.
Jeremy Chang: From Taiwan’s Perspective, Protective Tools May Not Conflict with Diffusion
During the panel, DSET CEO Jeremy Chih-Cheng Chang responded from a “Taiwan perspective,” noting that Taiwan’s role in global supply chains over the past three decades aligns with the importance of diffusion. Built on hardware manufacturing and supply-chain ecosystems, Taiwan has absorbed technologies, scaled production, and delivered them to global markets—developments that have not only enabled TSMC’s rise but also cultivated extensive local supplier networks and talent.
Chang added that while Ding uses “Fortress America” to critique protective policies such as export controls, research security, and investment screening, Taiwan’s security governance and practical experience suggest these tools do not necessarily conflict with diffusion. In his view, they can serve as risk-reduction mechanisms that help sustain the accumulation of critical capabilities even as democratic partners pursue broader adoption and application.
Ding responded that trade-offs can indeed exist between “fortress-style” controls and “diffusion-first” strategies. He cautioned that if tighter research security generates spillover effects that erode the talent base and trust in the education system, it may ultimately undermine long-term diffusion capacity. On export controls, he argued that treating AI competition as a short-term sprint can push policy design away from a more consequential yardstick: long-term diffusion and productivity.
Liang-Rong Chen: “Leading Sectors” and “Diffusion” Are Interlinked—Platform Choices Shape the Mechanism, Not Just the Outcome
Liang-Rong Chen argued that AI should be understood as an extension of the computing revolution, and that today’s AI competition is better interpreted through the formation and diffusion trajectory of the PC industry. He noted that while the book describes Japan as initially leading in some digital products but failing to translate that lead into overall advantage, the more decisive factor may have been Japan’s early constraints in the PC race due to strategic choices.
Chen pointed to how the United States—through the Intel–Windows platform—helped shape an open architecture, accelerated standardization, and drove rapid, low-cost diffusion. As PCs became more affordable and easier to use, the user base and application market expanded, drawing engineers into software development and fostering a self-reinforcing diffusion cycle. By contrast, Chen argued that Japan’s comparatively closed, independent-system orientation made it harder to replicate diffusion at similar speed and scale. He also noted that while the book treats “leading sectors” and “diffusion” as separable—suggesting a country could lead in one but lag in the other—PC history shows the two are tightly linked: platform-driven mass adoption is a key mechanism that initiates diffusion and subsequently fuels innovation and talent clustering, rather than merely a result that appears after diffusion.
Ding responded that his focus is on whether a GPT can be adopted more quickly and broadly across sectors and converted into productivity gains. He suggested that the United States advanced more comprehensive computerization across industries in part because it developed a relatively more effective model for training software engineering talent and supplying human capital. Japan’s tendency to concentrate resources in a small number of “centers of excellence,” he argued, may have constrained the speed and breadth of cross-industry diffusion.
Cheng Ting-Fang: U.S. Chip Policy Pressures Allies, and Blacklist Effects Are Hard to Reverse
Cheng Ting-Fang approached the discussion from Washington policy dynamics and allied responses. She asked how the United States should adjust its AI policy under a “marathon” framework, and highlighted how export-control signals (including developments related to the H200) and policy uncertainty create spillover pressures for democratic allies such as Taiwan, Japan, South Korea, and Europe. She cited Taiwan’s recent use of NVIDIA H200 chips to build an AI supercomputer as an example of how shifts in U.S. policy can have immediate regional implications.
Cheng also raised a longer-term, more pragmatic question: given that large-scale export controls and blacklist effects are difficult to reverse—and that China’s localization drive continues—would a U.S. shift toward a “diffusion-first” policy lens still be able to materially reshape the structural trajectory of competition?
Ding responded that both the United States and China often invoke “diffusion” in policy rhetoric, but there remains a gap in implementation and resource allocation—especially in the foundations of education, research support, talent development, and international student policies. He warned that constraints in these areas would weaken the “GPT skills infrastructure” that underpins long-term diffusion, while policy unpredictability itself raises costs. On export controls, he emphasized that effectiveness must be evaluated against policy objectives: if the goal is long-run productivity and national power, export controls may not change outcomes on a 20–30 year horizon and could produce counterproductive effects. By contrast, he argued that strengthening talent pipelines, education systems, and society-wide capacity to absorb new technology remains a more resilient priority—one less vulnerable to short-term policy swings.
Further reading: “Technology and the Rise of Great Powers: Innovation Does Not Equal Strength—AI as a General-Purpose Technology and Diffusion as the Key to Rebuilding Economic Power” (in Mandarin)
