“AI race” between the United States and China has permeated public discourse.
But the idea of an AI race between China and the US isn’t grounded in reality. The researchers, companies, and governments behind Chinese and US AI development are pursuing completely different goals.
The discourse in the US assumes that achieving artificial general intelligence (AGI)—computers that mimic human consciousness—would be so momentous and earth-shattering that clearly this must be the goal of anyone pursuing AI development. But that’s not the main goal of Chinese AI development. And a competition in which the competitors are running toward different finish lines isn’t a race.
Yes, China is developing LLMs, although largely in an open-source way as opposed to the for-profit competition in the US. Recent news stories report that China is “catching” the US in LLM development. Indeed, the latest Chinese model outperforms leading US models. But this isn’t evidence of an LLM-AGI race. Instead, it shows that without making AGI its main focus, China is capable of developing its own models almost as quickly as US companies.
Maybe we don’t need to “beat them”, but we do have to have an adequate industry of our own, lest the Chinese (or any other country, although thee are none at the moment) take hold of this vital industry the way they have with rare earths, display screens, solar panels, and soon, EVs.
I don’t think you want US data residing on servers in China, and I’m sure the military - which will have uses for AI, perhaps over and above those in many civilian sectors - doesn’t want that.
It’s not clear to me yet that AI will have the same kind of network effects that the telephone, internet, and similar have had - but I haven’t really thought about that part of it that much. If it does and it’s a “winner take all” technology, then that’s a problem (to be countered by making sure you have at least a strong second position). If it’s not, then we can go on our merry way, whistling through the graveyard as though nothing is wrong, and we can trust our future to the overlords of tech, who have our best interests in mind, always.
I need to be careful with what I say about such topics. Besides, I truly have very little insight into this. But I would say, as a lay person, that perhaps the race itself isn’t really bogus. It might not be different finish lines after all, but rather radically different approaches to the race itself.
From a car racing analogy consider that it might not be a race between two cars on the same track, where each car is radically different from the other. Rather, it is two tracks, with two separate starting lines, but they join at a common finish line.
Using that analogy, I think there’s two different finish lines. The U.S. industry and the Chinese industry are trying to achieve two very different things, for the most part.
The U.S. hyperscalers are trying to build AI God (to borrow Matt Levine’s term). They’re building frontier models that aim to create artificial general intelligence. Something that is capable of generalized intelligent behavior, and can thus do most anything that a human person can do.
Chinese firms are trying to build specialized AI’s that can do economically useful things without having to have general capabilities. They can code or manage your calendar or whatever, but they can’t do everything a frontier model can.
So the Chinese build stuff that’s not going to be a step change in intelligence, but is far less expensive, and therefore more commercially viable. The U.S. frontier model builders pursue that moonshot of an actual intelligence. Because our hyperscalers have (for now) been liberated from the need to show how their products will actually pencil out, under the assumption that the AGI will be so insanely valuable that cost is (or should be) no object. But for China’s lesser ambitions, the lower cost is sufficient to justify the more modest objectives.
That is happening in the US as well in fields from oil exploration to ag to drug discovery. When I’ve brought up these sort of things here they are typically dismissed as ‘not really AI’.
How good the LLM is by a little bit more at times here or there the US or China matters little. The Cloud right now is necessary. Meanwhile China makes appliances of it. We are turning to the PC and Mac for AGI.
Think without the hardware, AI does very little work.
The United States AI investment is massive, representing the largest concentrated technology spending boom in modern history. Total global AI spending is projected to hit $2.5 trillion****, and the U.S. commands the overwhelming majority of that capital. [1, 2]
U.S. spending on artificial intelligence and data center infrastructure is on track to consume roughly 2% to 2.5% of total U.S. GDP. This economic allocation matches or exceeds the country’s entire higher education sector and heavily rivals the scale of the national defense budget. [2, 3]
1. Private and Venture Capital Investment
Unparalleled Scale: U.S. private AI investment reached $285.9 billion. This dwarfs the rest of the world, pulling in 23 times more private capital than China and nearly 70 times more than India. [4, 5, 6, 7]
Mega-Rounds: Private funding is highly concentrated in foundational, general-purpose AI platforms. Record-breaking capital raises—such as OpenAI’s $40 billion round and Anthropic’s $13 billion round—greatly elevate U.S. venture activity. [8, 9, 10]
Startup Ecosystem: The U.S. leads globally in startup creation, with nearly 2,000 newly funded AI companies, more than 10 times the volume of the next closest nation. [5, 11]
2. Corporate Tech Giants (Hyperscalers)
Trillion-Dollar Track: The primary engine behind U.S. spending comes from “hyperscalers” like Microsoft, Amazon, Google, and Meta. Wall Street estimates project that annual capital expenditures from these few companies alone will breach $1 trillion by 2027. [12, 13, 14]
Infrastructure Bottlenecks: The capital is largely being funneled into real estate, green energy grids, advanced liquid-cooled data centers, and massive clusters of Nvidia chips. [1, 2, 12]
3. Government and Defense Spending
Federal Contracts: Public sector investment is escalating dramatically, primarily via the Department of Defense. The potential value of federal AI contracts climbed to $90.7 billion.
Legislative Backing: Beyond direct procurement, the U.S. continues to deploy hundreds of billions toward domestic semiconductor manufacturing and science research under the CHIPS and Science Act. [1, 15, 16, 17]
Historical Context: How Big Is This Boom?
To put the scale of current U.S. AI spending into perspective, economists compare it to history’s largest humanity-defining mega-projects (all figures adjusted for modern inflation): [18]
The Manhattan Project: $36 billion
The Apollo Program (Moon Landing): $250 billion
U.S. Interstate Highway System: $620 billion
U.S. AI Infrastructure Spend (Single Year):$650 billion to $800 billion [1, 2, 3]
Would you like to look closer at which U.S. tech giants are spending the most capital, look into the risks of an AI financial bubble, or see where the U.S. government is focusing its defense AI contracts?
Carefully examine the following numbers. You will find the EU per capita is spending more than China.
No, the EU is spending significantly less than China on data center construction and infrastructure. While Europe is attempting to scale up its digital presence, it faces a severe investment deficit, higher operational hurdles, and smaller capacity limits compared to China. [1, 2, 4]
Economic data highlights that China spends roughly 3.5 times more than Europe on data center construction heading toward 2030. Consequently, China currently commands 1.5 times the total data center capacity of the European continent. [1]