In a wide-ranging interview first published in the South China Morning Post, Jeffrey Ding, an assistant professor of political science at George Washington University and author of the award-winning book Technology and the Rise of Great Powers, argued that the true battleground in the US-China AI competition is not who invents the next revolutionary model, but which nation can spread AI most effectively across its economy. Ding, who also runs the ChinAI newsletter tracking China’s AI industry, said the US is “training for the wrong race” by focusing too heavily on an “innovation sprint” toward artificial general intelligence (AGI) rather than the slower, broader “diffusion marathon” that determines long-term national power.
Ding’s argument rests on a historical view of AI as a general-purpose technology (GPT), similar to electricity, which only delivers transformative productivity gains after years of widespread adoption. He told the South China Morning Post that the US and China are competing not in a short dash toward AGI, but in a decades-long effort to integrate AI into businesses across regions, industries and income levels. “Diffusion matters more than technological innovation for determining overall national power,” Ding said, noting that the Soviet Union once led in electrical innovation but failed to match the United States in spreading the technology through its economy.
The professor stressed that many analysts and policymakers focus on innovation metrics—research spending, top university rankings, number of highly cited papers, or frontier AI labs—while overlooking diffusion indicators. He suggested a more meaningful measure is how well a country can transfer cutting-edge breakthroughs from elite institutions into small and medium-sized enterprises in places like China’s inland provinces or the US Midwest. This, he said, is where AI’s real geopolitical impact will be determined, because economic growth and productivity gains ultimately shape military and strategic strength.
Ding was sharply critical of the Trump administration’s AI policies, calling them “counterproductive” despite appearing to emphasise diffusion. While the administration’s AI action plan acknowledged that slow adoption is a major bottleneck, Ding argued that other actions have undermined the United States’ long-term diffusion capacity. He pointed to the administration’s stance against the broader higher education ecosystem, its restrictions on international students, and a weakening of workforce development initiatives, all of which damage what he calls the US “GPT skill infrastructure.” “The lasting damage…is the disinvestment from education initiatives that help social mobility, support community colleges and the base of higher education institutions,” he said.
The interview also addressed the role of export controls on advanced semiconductors. Ding argued that these measures are rooted in a “fortress America” mindset focused on preventing Chinese access to cutting-edge technology. But he said such controls are not aligned with a diffusion marathon strategy and may harm US companies like Nvidia without yielding meaningful long-term advantage. “By the time AI has made a meaningful difference to the trajectory of national economies, export controls will have already eroded Nvidia’s monopoly on the most cutting-edge chips,” he said.
Turning to China, Ding challenged the prevailing view that Beijing is better at diffusing AI than innovating. He said China faces a “diffusion deficit,” where innovation capacity outpaces the country’s ability to spread technology broadly across the economy. He highlighted that China’s AI strengths are often measured by innovation indicators—top researchers and frontier models—while diffusion indicators show a different picture. For example, Ding tracks how many universities in each country have published at least one AI conference paper, a metric where the US far outpaces China. “Once you lower the quality baseline beyond the top 1 per cent of researchers…you find that the US still has an advantage,” he said.
Ding also criticised China’s highly centralised AI education push, arguing that top-down policy can become quickly outdated in a fast-evolving technology landscape. He said the country’s focus on AI majors and centralised planning may not be as effective as a more decentralised approach that fosters strong links between academia, industry, and smaller businesses. He also highlighted that China’s public education spending as a share of GDP is lower than other newly industrialised nations, while R&D spending is comparatively high, suggesting an imbalance in priorities that may undermine long-term diffusion.
On the question of whether AI will meaningfully boost productivity before 2030, Ding remained sceptical. He said the timeline for widespread diffusion is likely to be much longer than many public narratives suggest, and that the current hype could resemble a bubble. Even so, he maintained that the US remains well positioned to win the diffusion marathon due to its broad network of educational institutions and its deep talent base, though he warned that the US could lose ground if it continues to undermine the foundations of its AI diffusion capacity.
Ding said that measuring AI diffusion remains an open question, and that the most useful indicators may involve tracking how many small and medium-sized enterprises achieve significant usage thresholds for AI models. He also highlighted open-source software as a potential diffusion accelerant, noting that vibrant ecosystems can enable broader adoption, even if the leading-edge models are developed elsewhere. “Who cares where electricity comes from? What matters is that you can adopt it and get productivity gains from it,” he said, summarising the core message of his long-term view of the AI race.
