Editor’s note: I’m in the habit of bookmarking on LinkedIn and X (and in actual books, magazines, and newspapers) things I think are insightful and interesting. What I’m not in the habit of doing is ever revisiting those insightful, interesting bits of commentary and doing anything with them that would benefit anyone other than myself. This weekly column is an effort to correct that.
With AI infrastructure now a geopolitical battleground, is it really an open-ended tête-à-tête or is there a point where the winner takes most?
Perhaps macroeconomic uncertainty around on-again/off-again tariffs imposed by the United States is amplifying discourse but in the modern world, technological dominance is a key influence on global power dynamics. Prior to the AI explosion we’re in currently, all eyes were on semiconductor fabrication, a head-to-head battle between the US and China with Taiwan caught in the middle. Now with AI the focal point — and the enabling chips largely manufactured in Taiwan — the fight continues. And the big question is: is there a point in the future development of AI where a lead becomes so durable that it informs long-term hegemony?
A few months ago, Chinese startup DeepSeek released a head-turning model trained on a fraction of OpenAI’s budget that performed comparably in many tasks. The takeaway was that Chinese firms are capable of competing despite constraints. That said, most frontier model breakthroughs take place in the U.S.
More recently, U.S. export restrictions targeting NVIDIA chips have created another bottleneck. Previously, NVIDIA’s R20 GPU could still ship into China, but U.S. Department of Commerce export controls have blocked that path. In response, Huawei has pushed forward with its Ascend 910, aiming to provide a homegrown alternative. Again, it’s a self-reliance play that foregrounds China’s capabilities in the face of constraints.
Is AI really an “infinite game”?
At the recent Hill & Valley Forum in Washington DC, NVIDIA CEO Jensen Huang downplayed the idea that China is materially behind the US in AI. In fact, he noted that “50% of the world’s AI researchers are Chinese.” That should “play into how we think about the game.”
Huang called Huawei “one of the most formidable technology companies in the world…incredible in computing and network technology, all these essential capabilities to advance AI.” More broadly, “We are very close,” he said of the competitive dynamic, seemingly borrowing from game theory, referring to it as “an infinite game.”
American academic James Carse’s book Finite and Infinite Games was published in 1986, and it has become a logical construct used to inform business strategy that’s notably evangelized by Simon Sinek; he penned a book expanding on Carse’s thinking called The Infinite Game. The idea is that finite games have known players, fixed rules, and an agreed upon objective. Infinite games have known and unknown players, fungible rules, and the objective is to perpetuate the game.
Sinek’s thinking is that there is no winning in things like business or education or politics. And problems arise when a finite game mindset is applied to an infinite game. This foments “the decline of trust, the decline of cooperation, and the decline of innovation,” according to Sinek.
Hardware overhang may be a strategic advantage
But what if this AI arms race isn’t finite or infinite, rather some kind of complex, unstable hybrid game? Like a finite game, it has known players. Like an infinite game, the rules keep changing. But as the game progresses, if one player gains a durable advantage in compute and networking infrastructure scale, and model performance, things China appears to be quite adept at, couldn’t the game effectively end with the winner taking home durable geopolitical advantage?
There’s this idea that’s been floating around in the (online) discourse about how transformative artificial intelligence (TAI) or artificial general intelligence (AGI) or some hugely impactful inflection point is achieved. Open Philanthropy’s Ajeya Cotra takes this on in the very interesting 2020 paper, “Forecasting TAI with Biological Anchors.” And this is particularly resonant five years later as American heavyweights AWS, Google, Meta and Microsoft have committed to $315 billion in 2025 capital investment.
Cotra looked at the idea that “AI progress is ‘hardware bottlenecked,’ in the sense that the main (or only) factor limiting further progress in AI capabilities is progress in the availability of hardware.” But there’s a “controversial” contra-position — one that DeepSeek’s emergence should have highlighted — “that progress is bottlenecked by key algorithmic insights, and dramatically increasing hardware without acquiring those insights would have very limited value.”
To restate that, the predominant thinking is that we need more AI infrastructure for AI to revolutionarily, not evolutionarily, advance. But maybe we’re just waiting on the right bit of algorithmic alchemy. Expanding on that idea, if the US or China reached some point of software breakthrough, the spoils would go to the country that has the hardware overhang to diffuse it the most quickly.
As AI evolves, CSIS sees a “missing link” in the US
In their recent article, “The Missing Link in the AI Stack: Why Digital Infrastructure Is Essential to US Leadership,” Navin Girishnakar and Matt Pearl of the Washington DC-based Center for Strategic and International Studies think tank explored a “key enabler of the AI stack, and it is one on which the United States is vulnerable: the networks over which AI traffic travels.”
The co-authors call out trends around the commoditization of AI models (with DeepSeek as the case study), and the importance of the network edge as both a place where AI systems acquire data and where AI applications touch end users. “For AI to flourish, the United States will need communications infrastructure…of substantially greater density, complexity, and scale. This infrastructure will need to be multimodal, reaching every edge device that can potentially leverage AI applications.”
While China is actively investing in networking via domestic companies, Western firms are “individual market participants…The United States is in desperate need of a comprehensive strategy to bolster digital infrastructure.”
I conceptually don’t like binaries; the world is far from black and white. That’s born out in post-digital technologies, especially in quantum computing. The fundamental digital computing logic of 1s and 0s gives way to superposition where there are 1s, 0s, and both at the same time. Maybe that logic also applies to AI in that it’s not a finite or infinite game, but something structurally different, something hybrid.
Whatever the game is, AI infrastructure is vital
As it relates to AI infrastructure, more is more. If hardware is the constraint, keep on building. If algorithmic advancement is the constraint, but the advantage goes to whomever can scale faster, keep on building. Either way, former Google CEO Eric Schmidt and his office’s China, AI, Policy Research and Strategy Lead Selina Xu summed it up correctly in a co-authored guest essay for the New York Times.
“China is at parity or pulling ahead of the United States in a variety of technologies, notably on the AI frontier,” they wrote. Schmidt and Xu called out China’s position in electric vehicles, robotics and STEM education. “The China-dominated future is already arriving — unless we get our act together.” That’s not to say that the US doesn’t have some advantages and China certainly has its own constraints.
However, “We’re no longer in the era when China is far behind us,” Schmidt and Xu concluded. “If China’s capacity to innovate endures…then the next chapter of the AI race will be an all-out dogfight on every axis possible. America will need every advantage it has.”
For a big-picture breakdown of both the how and the why of AI infrastructure, including 2025 hyperscaler capex guidance, the rise of edge AI, the push to AGI, and more, download my report, “AI infrastructure — mapping the next economic revolution.”
