From the South China Morning Post, January 26/February 4:
Professor says a ‘diffusion marathon’, where AI is gradually embedded in an economy over decades, will determine the winner of US-China tech war
Jeffrey Ding is an assistant professor of political science at George Washington University. He is the author of Technology and the Rise of Great Powers, an award-winning book exploring the impact of technology on geopolitical competition, as well as the founder of the ChinAI newsletter, which tracks developments in China’s artificial intelligence (AI) industry.
In this interview, Ding explains why “diffusion”, not innovation, will determine whether China or the US will prevail in the AI race, the Trump administration’s “counterproductive” policies around the technology, misconceptions about the two countries’ respective strengths in the field and why human capital is the key to victory.
This interview first appeared in . For other interviews in the Open Questions series, click .
You recently wrote an article for Rand Corporation, the influential US think tank, in which you argued that the US is “training for the wrong race” in AI. What did you mean?
The main reason I wrote that piece was to clarify what I see as a lot of confusion out there about what the US and China are actually competing for in AI. Others have already articulated different visions of this US-China AI race, but I wanted to put forth clearly that there is one type of race that I think the US should optimise for, which is this “diffusion marathon” rather than a sprint towards a clear finish line.
The “diffusion marathon” refers to the progress the two countries make in spreading AI throughout their respective economies.
This can be contrasted with a vision of the AI race as an “innovation sprint” – the view of many in US national security circles – where the key question is which country can innovate its way to developing an artificial general intelligence (AGI) with “God-like powers”, in the words of Jake Sullivan, the National Security Adviser under the Joe Biden administration.
In their view, this sprint will be decided in the next few years, and will give the winning country a permanent technological lead and decisive military advantage, whereas the “diffusion marathon” that I am putting forward happens over decades, with no obvious finish line.
Your book argued that diffusion matters more than for determining overall national power. Why is that?
It’s because AI is a general-purpose technology (GPT), something that can spread throughout an economy, rather than just be limited to certain industries or sectors.
If you look at the examples of previous GPTs such as electricity, they stand out as historical engines of growth because they boosted productivity and transformed entire economies. And history tells us that a country’s ability to sustain economic growth in the long run is central to its geopolitical and military influence.
The difficulty for policymakers is that the potential economic gains from a GPT will arrive only after a long process of diffusion.
For example, while the Soviet Union was an innovation leader in electricity, it was the US that became the biggest diffuser of the technology by 1970. Even then, the boost to US productivity from electrification was only seen four decades after the first electric dynamo emerged, because complementary innovations such as steam turbines had to be made for these gains to be realised.
When many journalists and policymakers think about scientific and technological power, they often focus on a set of indicators revolving around innovation capacity: a country’s ability to pioneer new-to-the-world inventions. For example, how much is a country spending on cutting-edge research and development? What is the average international ranking of your top three research universities?
Specifically for AI, the indicators people tend to focus on are top AI start-ups or which AI labs or institutions have published the most highly cited papers that year.
However, there is another set of indicators to measure a country’s scientific and technological prowess, what I call diffusion capacity. There, it’s not necessarily about the frontier AI companies or universities, your Stanfords or Tsinghuas, or your OpenAIs and DeepSeeks.
Rather, it’s about which country can take the technological breakthroughs incubated in these frontier institutions and transfer them to small and medium-sized businesses in an inland province like Qinghai, or a Midwestern state like Iowa where I grew up in the US.
This is the most important part of US-China AI competition, based on years of historically grounded research into past industrial revolutions and GPTs.
What is your assessment of the ? It appears to have placed more emphasis on AI diffusion than the previous Biden administration, which seemed more focused on sprinting towards AGI.
It’s been interesting to see how so many people are now paying attention to AI diffusion, which was a very neglected topic back when I started writing my book.
The Trump administration’s AI action plan does gesture at some of these components of the diffusion marathon. For example, it says that the slow adoption of AI is the key bottleneck holding back the technology, not necessarily a lack of cutting-edge AI models or applications.
But my view is that this is akin to treating an ankle sprain while the patient is bleeding out. Overall, the Trump administration has done much more damage to the US’ ability to diffuse AI over the next few decades than it has helped....
....MUCH MORE
Here's a comment on diffusion introducing November 2020's "Investor's Business Daily on Artificial Intelligence Stocks":
There is a definitional problem with the term "AI stocks [or companies]" in that AI is a tool. Much as the (over) hyped nanotechnology revolution didn't produce "nanotech stocks" but instead became incorporated into processes and procedures that give companies employing same an incremental edge rather than epochal shifts.*
As noted in the outro from a 2017 post:
...Much more important than the direct monetization of big data is the strategic advantage it can bestow over time.
In a winner-take-all economy, as in a horse race, small differences in superiority are rewarded all out of proportion to the actual advantage. A top thoroughbred may only be a couple fifths of a second faster than the field but those two lengths over the course of a season can mean triple the earnings for #1 vs. #2.
In commerce the results can be even more dramatic because rather than the 60%/20%/10% purse structure of the racetrack the winning vendor will often get 100% of a customer's business.
However, if there is an AI "company" Nvidia would deserve the moniker as much as anyone.
If interested there are quite a few back-links in November 2025's "Robots and AI Are Already Remaking the Chinese Economy".
