With the expected money and influence that is to come from being an AI-powered country, there is also an expectation of great geo-political leverage. Specifically, Artificial General Intelligence (AGI) is a massive target that is constantly talked about. This technology would provide both hard and soft power. Not just from the cash flow that comes with its adoption by a country and its businesses, but also the extent of adoption of your particular LLM around the world. There are lots of political consequences that come from countries creating sufficiently powerful and intelligent AI, of which I will discuss in this blog.
Artificial General Intelligence does not actually have one strict definition. But it is usually used to refer to AI that is at least as intellectually capable as humans in all fields. This is the threshold at which in all aspects it is more efficient to delegate work to AI than humans. I think it’s actually an odd label, because it implies that at some point in the development of AI, a switch suddenly turns on past a certain threshold where you get the AGI tick of approval. In reality, the development of AI has been a continuous, gradual exponential uplift. The point at which it becomes AGI is, in my opinion at least, arbitrary. It sounds well defined since you have a comparison to humans, but what does “everything” entail when we say it must be better than humans at everything? How do we measure better? What if AI becomes extremely transformative before hitting AGI because it’s better than humans in enough dimensions and by enough of a margin?
Demis Hassabis himself has defined AGI as AI that is capable of outperforming every human at every task. There was then a debate with Demis and Yann LeCun, where LeCun says even humans are not examples of general intelligence because the intelligence is specialized for survival, whereas Demis replied saying that is universal intelligence, which is different to general intelligence. Dario Amodei has said he does not even like the term AGI, and prefers the term “powerful AI”, keeping a vague label to match the vague semantics and enforce the idea that timelines and projections of future AI capabilities are ultimately extremely speculative. Karen Hao, a journalist who has written a very detailed expose of Sam Altman and OpenAI says AGI is not even something that will come, and it is just used to push a narrative to achieve an end goal of extreme wealth and power. I see it as no surprise that the discussion on defining AGI turns into a debate of semantics, given the initial premise of AGI (in my opinion) largely stands on flimsy semantics in the first place.
Regardless of what it actually means, it became an incredibly popular buzzword and is very useful for framing to the public. When the public want to know what the future will look like and how soon, “how long until AGI” is an incredibly simple and tangible way to do so, even if ill-defined. As Karen rightly points out, it also makes an easy narrative for the competition framing against China. Needing to reach AGI before our enemies is an extremely easy story to understand for the public, and an easy way to pull support from the patriotism of citizens, money from VCs, interest from the government and a sense of urgency from everyone. While I think Karen overstates this point, I do think it is a valid one. But why do we need to beat China? Can the world not work together on this given the stakes at hand?
It seems incredibly unlikely the world will be working together fully. Though there is limited collaboration, with research papers and open weight models generally being published publicly. When Chinese company DeepSeek shocked the world with a cutting-edge model, both their model weights and their research was made publicly available! Likewise, papers have been publicly released after big LLM releases from western companies too. But generally speaking, there is just too much to gain from your country producing the most advanced AI technology to incentivize competition rather than collaboration, as is the design of capitalism.
The gains are money, power and influence, that old infamous trio. Given the productivity gains which AI can provide, the global market for AI is enormous and will only continue to grow. If a country is successful in solidly pinching a large portion of this market share, there would be enormous gains in wealth for this country. Suddenly, you have influence over the information fed to anybody using your LLM. Imagine, alongside the huge potential for these general chatbot assistants to educate people who have no other means to access of teachers or even information, if these chatbot assistants also hold your values while teaching these people. For example, Chinese models already hold socialist values and censor anything negative about the Chinese government, as mentioned in the last blog. This gives a lot of soft power to countries who can hold on to significant market share. If you have the most powerful AI technology by a sizeable margin, naturally more people will use your product, and your ability to shape the view of citizens overseas and potentially even influence the way citizens of countries think, even if to a limited extent. Now, in the context of this West vs East framing, the problem makes more sense. There is a race to influencing the world on values that are conducive to freedom of thought and democratic values vs values that indoctrinate citizens in a way that is advantageous for autocratic power. Though there is always a risk of democratic powers trying to do what we are afraid of autocratic powers doing, generally speaking this West vs East framing is about freedom and democracy prevailing.
While there is a valid aspect to this framing, there are definitely financial incentives for pushing this narrative as well. It’s the perfect story to get people to sympathize and for VCs to send money. The characters and storyline go like this: There are the good guys (the west), the bad guys (the east) and high stakes (democracy and freedom of thought). The line between genuinely wanting these values enforced and utilising this story for the success of your own company can be blurred, and I think the extent to which this two-faced aspect plays a part does vary from company to company.
There is currently a big debate on whether the US should export Nvidia chips to China. All mass production of cutting-edge GPUs take place in the US with Nvidia being a US company. GPUs are the currency to compute, which in turn is the currency to the development and building of AI. Compute and consequently GPUs are the biggest bottleneck to AI development, and China is completely reliant on the USA for chips, a big strategic play for the US is to stop exporting chips to China. This would force China to make their own chips, thus slowing them down. So, the pro is buying the US time to get ahead of China. A con is that this leads to a loss of income from China buying these chips. Another problem with this plan is that China will eventually develop GPUs on par with the US, at which point the US will have lost the money it is getting from China for purchasing these GPUs permanently, and the US loses the leverage they get from China relying on them for GPUs.
Currently, the US has several companies which produce frontier models. OpenAI’s ChatGPT, Google’s Gemini, Anthropic’s Claude and Meta’s Llama, the whole of which holds a significant market share of the LLM market. In China, they have DeepSeek’s DeepSeek, Z.AI’s GLM, Alibaba’s Qwen, and Moonshot’s Kimi. Some of these companies made it to the frontier of LLM technology very recently. Both Kimi and DeepSeek have been frontier models for about a year and Z.AI released their first frontier LLM GLM 5 February 2026, extremely recently! Given China had no frontier LLMs before DeepSeek last year, we do not have much information to go off for whether China can sustain their position as a serious competitor, but my gut instinct is very likely yes.
In Europe, Mistral AI is the most prominent LLM though even this is not considered a frontier model. Due to European laws being much stricter on AI regulation, this has largely inhibited the ability to keep up with the pace of progress seen in the US and China. At least for now, Europe is not a main player.
Both Saudi Arabia and the UAE are currently building data centres (more on this later) with the long-term strategy of being a big player in AI. Saudi Arabia have a company HUMAIN which is backed by the state and the prince of Saudi Arabia itself! Using the vast amount of wealth on top of the vast energy stores they have in oil, they plan on investing lots of money into AI. They also have a partnership with American chip company Groq to buy the resources needed to build the needed infrastructure. OpenAI also announced a partnership with Groq and HUMAIN to host their open-source models using Saudi Arabian data centers. Elon Musk’s xAI has also entered a large deal to utilise these data centres for Grok (Groq is the chip company, Grok is the LLM created by xAI, confusing!) Given these countries are dictatorships, naturally a similar fear to China arises with these countries. Interestingly, using their data centres is very much in contrast to the harsh restrictions on the US exporting GPUs to China! Will these models include strategically biased viewpoints, censorship and will they be used to repress their citizens rather than empower them? At the moment this is actually not the case! While there are no frontier LLMs from these countries, the existing ones typically do not steer towards certain viewpoints in the way Chinese models do with socialism and other CCP narratives, though there are reports of these models avoiding sensitive topics, political and religious in nature. My hunch is they are a middle ground between the no-steering philosophy of the west and the heavy-steering philosophy of China. As a side note, this is not to say there is no bias in western models to western values, but this bias is not an intentional feature, but an unfortunate byproduct of training on mostly English text that will naturally contain western values.
The massive upfront cost for training frontier models really sets the stage for how these dynamics work. Since all the big tech companies are in the US and Silicon Valley is in the US, this naturally led to the early lead the US took (though, as mentioned earlier, China’s trajectory is incredibly uncertain as of now given how recent their come up was). Big tech companies have the perfect circumstances for AI since they have both the capital, technical talent, and pre-existing brand name needed to hit the ground running with these models. They also have the pick for the top talent from top universities, or from wherever else top talent comes from. You won’t see a random frontier LLM come from a developing country unfortunately, or any useable LLM for that matter due to the vast amount of resource needed. Who knows, maybe as the technology gets better, they will become more accessible! Though this could very well never be the case if off the back of increases in efficiency, there is never an incentive to cut back on used resource due to hitting a limit on AI capability where we don’t need to use more resource/compute and can direct the efficiency gains to lowering resources needed to make a model at this “capped” level.
I have mentioned this here and there in previous blogs as well as this one, but China is very likely going to use the capabilities of AI to control and spy on their population given they already do this to the best of their ability with the technology available to them. If you had a camera on every street corner and even in every house, you could use AI to sift through the mass amounts of video and audio and flag any behaviour which is not to the liking of the government. Then you could use AI to identify any people that have been flagged and match them to a massive database of every citizen in China. Then the government can find and do what they deem necessary to anybody flagged. This would go beyond a police state, to an Orwellian reality.
In the US, while the government has been looking for a partnership with one of these major AI companies to adopt their technologies within the government, there ended up being some controversy with Anthropic, who they initially were looking to make a $200 million deal with. The deal fell through due to an inability to come to agreement with the terms of the deal. The US government wanted to use the technology for “any lawful use” which Anthropic were not happy with, since this was a way to bypass the moral boundaries established by the company on building autonomous weapons and, what I will now be discussing, mass surveillance of US citizens.
The US government has a deal with Palantir to use their technology, and while it’s not 100% clear what Palantir does exactly, it is known that they have a comprehensive database containing US citizen data. Similar to what I said about China, if the government have unrestricted access to these language models, mass surveillance becomes very possible. Given they were trying to circumvent the red tape put down by Anthropic on autonomous weapons and mass surveillance, this suggests they wanted the freedom to indulge in these activities.
In the end, this deal fell through and OpenAI ended up making the deal instead. On top of this, Donald Trump tried to ban the use of Anthropic models for all government agencies, and the Department of Defence (DOD) blacklisted the company on the grounds of Anthropic being a “supply chain risk”. Anthropic have taken the DOD to court, and this legal case is still on going. This does not paint the government in a very favourable light in terms of respecting the freedom of their citizens. Trying to “punish” the company for not doing what they wanted also does not paint them in a favourable light. This is a sobering reminder that freedom and democracy does not levitate but is something to be propped up with continuous human effort.
As mentioned earlier, the middle east has been aggressively building data centres and making their own LLMs using their wealth and oil. If we focus on the data centres, going back to the deal OpenAI has made with HUMAIN to host their open-source models on Saudi servers, I think this point is very interesting. Relying on infrastructure and data centres in another country to train and host your models takes a lot of trust, especially with how powerful we expect AI to become in the future. For example, there have been deals for some data centres to be built in the UK, but of course the US and UK are very close allies with very similar values. These middle eastern countries on the other hand, whilst friendly, are very different to the US and the ties are not as close. The middle east becoming a large data centre hub utilised by the west could give these middle eastern countries leverage over the US. I am no lawyer and do not know the details of the contracts, but if there is a falling out and one or all of these countries decide to suddenly cut off all use from the US, or if they manage to “steal” model weights of the US models in western data centres, this could cause big issues for the US and I am not convinced the US could prevent this. This is very pessimistic of course, and we have no reason to assume this will happen. It is an interesting point to note though.
AGI, even if the definition is hazy, is extremely politically significant. This competition between the US and China to try and secure powerful AI capabilities for the world to use to prevent the other side’s agendas from permeating through the masses of the globe is incredibly high stakes, and neither side is going to let up easily. At the same time, while top tier GPU production residing in the US is strong leverage with the ability to cut off supply to China, whether this would allow for the US to gain ground or if it would just hurt the US economy from the lost revenue and force China to be self-reliant on GPU production, potentially successfully, is unclear. Despite this, some US companies like OpenAI and xAI have partnerships with companies in the middle east for using their data centres! They better hope Saudi Arabia do as the US says, and not as the US does. The aggressive building of data centres in the middle east using their wealth and oil energy is also a fairly new and very interesting development, and one to keep an eye on.
Though the US were dominating from the beginning, China just in the last year has really started to come out with what seem to be some very high quality LLMs (at least according to benchmarks). Given this, it seems that the race has just begun. The middle east also seems to be joining the competition, against themselves to form partnerships with other countries to use their data centres and potentially down the line, with the rest of the world if they manage to pull off creating a frontier LLM!
Meanwhile, with each country having their own incentives, you also have each individual LLM company trying to take as much market share as they can as well, leading to competition within borders as well as beyond, and ambiguity as to which of these two bodies of competition is on each companies mind as they push for this US vs China narrative. Those who prevail in the world of AI will be extremely rich. Maybe both have equal success, maybe one storms past the other, we will have to wait and find out where the scale tips and whether freedom of information prevails over censorship on the global scale.