A New Cold War, Rise of China and Global Chaos: How AI Will Rewrite Geopolitics | Gregory Smith
What this covers
Gregory Smith, a policy analyst at the RAND Corporation focusing on AI and geopolitics, discusses a framework for understanding how AGI or ASI could reshape the global order. The conversation centers on a recent RAND paper that maps eight scenarios across two axes: whether one actor or multiple actors achieve AGI first, and whether AGI empowers the US or its adversaries. Smith argues that AGI's geopolitical impact has been largely unexplored despite being potentially as transformative as nuclear weapons or electricity. The argument rests on a straightforward premise: since geopolitics is the sum of multiple national domains—military, economic, technological, political—and AGI would transform each of those domains, the technology must reshape the global order fundamentally.
The conversation moves through what distinguishes each scenario and which are most likely. Smith emphasizes that whether AGI behaves more like nuclear weapons (expensive, controllable, concentrated among a few states) or the internet (cheap, proliferating, easy to distribute) determines which future emerges. He identifies the Cold War II scenario as plausible—a bipolar competition between the US and China both possessing AGI—but notes that cost curves, reasoning models, and model distillation could make AGI cheap enough to undermine this outcome. He also addresses a crucial disanalogy: while the Cold War's nuclear standoff held because both sides understood mutual annihilation, AGI's risks are far more abstract, potentially lowering the psychological barriers to development and deployment. On major war, Smith argues it would likely be a deliberate choice rather than accident, since mutual AGI possession would not itself be a casus belli. Among the eight scenarios examined, Smith identifies three as most probable—Cold War II, the Authoritarian Advantage, and the Wild Frontier—and notes that all three present unfavorable outcomes for the US.
AGI and ASI development will likely transform geopolitics through one of eight distinct scenarios, primarily distinguished by whether one or multiple actors achieve AGI first and whether it empowers or disempowers the US relative to its adversaries, with most scenarios producing significantly destabilized global orders and shifts in great power dominance.
- AGI/ASI represents a general-purpose technology with applications across all domains, making its geopolitical effects a sum of transformations across militarily, economically, and politically relevant domains
- The two critical variables determining outcomes are monopoly vs. multipolar AGI development and whether AGI empowers or disempowers the US relative to adversaries
- Unlike previous transformative technologies, AGI's effects depend heavily on technical assumptions about cost, controllability, and speed of takeoff, creating radical uncertainty about which scenario materializes
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Mutually assured destruction in the Cold War held not because both sides were content with it but because neither side could technologically escape it, and both continually tried to escape it (first-strike theories, the missile race) until they realized they could not out-race each other.
“no one ... neither side really liked being subject to mutually assured destruction. It was ... a significant disadvantage to be in a mutual”
AGI is a general-purpose technology—'AI plus anything' (health, electronics, policy, military)—which creates strong incentives to deploy it very widely, making it far harder to control like nuclear weapons because it is fundamentally easy-to-proliferate software with large potential profits.
“AGI is a general purpose technology in that it can be like AI plus something else like add any domain AI plus health AI plus you know AI plus electronics”
The geopolitical implications of AGI/ASI have been substantially underexplored relative to analysis of impacts on other domains such as economics, medicine, and technology
“while the impact on other areas is widely discussed, how it might reshape geopolitics is largely ignored even though its impact would be absolutely transformational”
Destabilization risk concentrates in scenarios where one side getting to AGI first creates a decisive advantage (a fast takeoff), making the rival's progress itself a threat—this is where AGI sparking destabilizing relationships and potential armed conflict comes into focus.
“one side of their getting there is actually a threat to the other one because ... maybe you take off ... and all of a sudden you have a decisive advantage from being getting there first. And that's where AGI sparking destabilizing relationships comes into more focus”
The Cold War II scenario rests on a technical assumption that AGI development remains very expensive, so that large firms and capital-rich states (with talent and educational capability) have an advantage; if AGI development turns out to be cheap—via leveling cost curves, reasoning models that run longer rather than train longer, or distilling cheaper models from a frontier model's outputs—then fast-followers could proliferate the technology and undermine the bipolar scenario.
“building AGI is still ... expensive and therefore large firms and countries have an advantage ... if that's not true if AGI development turns out to be cheap”
AGI's potential impact is so significant that whether it arrives today, 10 years from now, or 50 years from now remains incredibly important and consequential, making timing less critical than acknowledging the magnitude of possible transformation.
“this tech you know AGI at sort of The way we looked at it is so impactful that if it shows up today, 10 years from now, 50 years from now, is still incredibly important and incredibly impactful. And that is both a blessing and a curse of thinking about the topic.”
Major war between great powers will most likely be a deliberate decision made for relatively extreme reasons rather than an accident; AGI—especially in the cyber domain—could be stability-deteriorating on a case-by-case basis, but barring something like a misread radar (as nearly happened with nuclear war), large-scale war won't simply be accidental.
“major war will probably be a decision rather than an accident”
AGI is a general-purpose technology that can be applied across any domain (health, policy, electronics, defense), meaning that even if AGI development is centralized, pressure to deploy it across all sectors will be enormous and difficult to resist
“AGI is a general per you know AGI is a general purpose technology in that it can be like AI plus something else like add any domain AI plus health AI plus you know AI plus electronics AI plus you know policy research uh and it will have an application. So even if both sides develop it, I expect there'll be pretty strong incentives to deploy it very widely.”
During the Cold War, mutual assured destruction (MAD) was theoretically coherent but both superpowers continuously tried to escape it through first-strike capabilities and technological advantages rather than accepting the logic of mutual vulnerability
“mutually assured destruction was recognized by both sides but both sides continually tried to escape it as a mentor of mine here at rand would often say like it wasn't like you know like people sort of like uh you know people you know their whole theories of like you know first strike you know first strike uh doing the you know all these other things where it's like oh can we like is it win it like can we win a nuclear war”
Defining AGI and ASI precisely remains contested across the field, with ongoing debates about what threshold constitutes AGI, what it means to perform all intellectually valuable tasks, cost-effectiveness of AI compared to human performance, and whether superhuman intelligence is possible.
“I will say defining AGI and ASI is sort of a you know everyone has their own definition of exactly what it means I'd say and at Rand there's um in particular and elsewhere you see a lot of debates about what AGI means like what does it mean to perform all those intellectually valuable tasks is that AGI again the question of can it get to a point where it can perform physical tasks without human as efficient as a human just because it can do it is it cost effective compared to a human”
The analogy between AGI and nuclear weapons (expensive, controllable, limited to few states) versus internet (cheap, proliferative, benefits from widespread adoption) is useful for understanding scenario variation but imperfect.
“I do agree that like it is like you know is it more like nukes like it is expensive easily control like easily like nucle nukes are compared to the internet very easy to control maybe not easy in some absolute sense but or is it more like the internet highly just built to proliferate basically which the internet has a lot of like like internet connectivity generally there's just a lot of reasons why it's valuable to proliferate it.”
Whether AGI produces a Cold War II depends on whether it resembles nuclear weapons—expensive, easily controlled, held by only a few states—or the internet—cheap and built to proliferate widely.
“it basically hinges on whether AGI will be more like nuclear weapons as a technology that only a few of the most powerful states have ... or if it's going to be like the internet that is very cheap and everyone has it”
Geopolitics is essentially the summing-up of multiple nationally relevant domains, so if AGI changes all those domains, the geopolitical order could change dramatically—reinforcing current trends, reversing them, or producing entirely new configurations of power.
“geopolitics is like the summing up of multiple you know nationally relevant domains and if all those domains are changing well the geopolitical order is going could change quite a bit”
Reasoning models are an innovation in which a model 'thinks' more—engaging an extended chain of thought and spending more compute at inference time before producing an output—rather than simply producing a single output, which shifts cost from expensive training toward longer runtime.
“reasoning involves a model thinking a lot more, engaging a chain of thought, spending a lot more time thinking and going over its thoughts rather than just making one output”
Reasoning models (models that spend more computational time thinking through problems rather than producing immediate outputs) could reduce AGI development costs if they rely more on inference-time compute than training-time compute, enabling cheaper model advancement
“reasoning models which are sort of an innovation where you have a model think a lot you know I will probably massacre this so to any technical folks listening to this I apologize but basically a model, you know, reasoning involves a model thinking a lot more, engaging a chain of thought, spending a lot more time thinking and going over its thoughts rather than just making one output before serving you an output. Um that is you know if if that if that methodology that method is mostly about having the com model run longer rather than training the model longer which and better which the training can be very expensive to do in singular. But if a lot of advantages can come from cheaper methods or having like a model run for longer on a small on a smaller computer.”
AGI development likely requires enormous capital investment, specialized talent, and educational infrastructure that only major countries with large existing tech ecosystems can currently marshal, making widespread global distribution of AGI development unlikely in the near term
“one of the things that this scenario requires or at least that we you know we hypothesize is that though there's multipolar development development is still very expensive building the next building AGI is still for one is still you know is still expensive and therefore large firms and countries have an advantage and you know they're the ones with the capital to get there who can marshall the resources talent educational capabilities to do so”
In a world where AGI then ASI moves quickly and is very impactful, getting there first can confer a decisive advantage, which is why the single-vs-multiple-developer axis was central—though there are also possible worlds where being first is not actually a large advantage.
“in a world where AGI than ASI ... moves quickly and is very impactful, getting there first can be very important”
AGI's geopolitical impact has been largely under-discussed and under-explored relative to its impact on other domains, even though within the next decade it could be extremely consequential and transformational for the global order.
“how it might reshape geopolitics is largely ignored even though its impact would be absolutely transformational”
The current US-China trajectory points toward the Cold War II scenario because they are the only two countries with sufficient resources and technological development, are in increasing competition, and in the US the China-competition argument is already used to resist slowing down AI development.
“China both China and the US are by far the two countries ... with the most or enough resources ... and at the same time they are in an increasing competition”
AGI or ASI that can perform all economically valuable intellectual tasks and significantly outperform humans in every field is a realistic possibility within the next decade
“we might get an AGI, artificial general intelligence or even ASI, artificial super intelligence, a next stage of AI that would be able to do everything that humans can do and possibly even significantly better. If that happens, it will radically transform every aspect of our lives.”
The paper deliberately avoids assigning probabilities to scenarios because the epistemic uncertainty about AGI development timelines, technical capabilities, and outcomes is too high to produce credible probability estimates
“you know I'd say on probabilities just that you know the team I think was more interested in understanding the potent potential futures. we you know the high level uncertainty just made it hard for us to feel we could really and you know you know you can assign probabilities to to these I think more re I think I think if you think deeply about it it's possible in the future too but that wasn't our objective here”
If you create something as intelligent as or beyond humans, there will necessarily be a very significant impact from its mere existence, regardless of the large uncertainty about the specific technical details, alignment, capability, or speed of takeoff.
“this is more a statement about if you create something that intelligent there's going to be a pretty significant impact from its existence with a ton of uncertainty about what those impacts are”
Once both superpowers possess AGI and are deploying it across their economies and societies, AGI is unlikely to be a spark for war in itself, because you would not go to war merely because the other side has what you also have; major war would more likely arise from some other specific trigger, with AGI making any resulting war more unpleasant.
“once both sides have it, I don't think they necessarily go to war over the fact the other one has it. Maybe something else sparks it and AGI enabled war is quite unpleasant, but it's not in of itself a spark”
Fully automating warfare with robots and drones could change escalation dynamics: states might destroy each other's expensive hardware without human casualties, making de-escalation easy (both sides 'go home'), so an AI arms race would create risks but the trigger for actually employing the weapons would have to be something relatively specific rather than the mere loss of machines.
“you shoot down my robot, you shoot down my robot. We all go home at the end of the day ... no one died. It's very easy to step back in that respect”
The paper deliberately defines AGI at its most capable extent—as intelligent as a human across all economically valuable or intellectual tasks—and ASI as far outperforming humans in every field, while bracketing the robotics/physical-embodiment question by assuming those physical problems are solvable.
“AGI is something that is ... as capable as a human intellectually across all the intellectual tasks humans perform or some equivalent”
The eight scenarios are structured along two key axes derived from expert interviews: (1) whether one actor reaches AGI first or multiple do, and (2) whether AGI empowers the US or empowers its adversaries (with adversary empowerment implying relative US disempowerment)—yielding a 2x4 framework rather than eight ad hoc ideas.
“one was sort of like is there one person who gets to AGI or they're multiple”
Cold War II is a scenario in which both the US and China achieve AGI and are transformed by it but remain in rough parity and bilateral competition; AGI tends to equalize the relationship (eroding some prior US advantages), coordination between the two sides proves hard, and competition remains the dominant frame without AGI obliterating the coherence of geopolitical rivalry.
“cold war II is very much a scenario in which both the US and China achieve AGI are transformed by it but remain in sort of a competitive ... environment of rough parody to each other”
The scenario where one side achieves decisive technological advantage before the other (fast takeoff) could be more destabilizing than bilateral AGI development because it creates a window where preemptive action appears attractive
“The idea that uh the idea that uh one side of their getting there is actually a threat to the other one because you know maybe you take off take off and all of a sudden you have a you know a decisive advantage from be getting there first. And that's where AGI sparking destabilizing relationships comes into more focus.”
The paper presents eight scenarios organized across two critical axes: (1) whether one actor or multiple actors achieve AGI first, and (2) whether AGI empowers the party that develops it or disempowers both parties simultaneously
“there were sort of these two axes in the paper sort of the uh one was sort of like is there one person who gets to AGI or they're multiple... The other was sort of, you know, does AGI empower the party who has it or does it um or does it or or not?”
AI-enabled military automation could reduce human casualties in military conflicts by replacing human pilots and soldiers with autonomous systems, creating a dynamic where both sides can engage repeatedly without incurring personnel losses unless explicitly escalated
“if you shoot I shoot down your robot, you shoot down my robot. We all go home at the end of the day. You know, I was piloting the ro, you know, I was piloting the robot from some from some like trailer in uh you know, in you know, in in Nevada and you were and you were piloting your drone from you know, some trailer in what in your prefer in whatever country you're from and it's like all right, we just go home at the end and we you know, both sides are very unhappy that they lost $6 million in equipment, but no one died. It's very easy to step back in that respect.”
Among the eight scenarios, Cold War II, the Authoritarian Advantage, and the Wild Frontier appear to be the most likely, and all three are bad—or differently bad—for the US, with none being particularly positive.
“the ones that I saw as the most likely were scenarios that you called War II, the authoritarian advantage, and the third was the wild frontier”
A key disanalogy between AGI and Cold War nukes is that with nuclear weapons both sides clearly understood the threat of global annihilation, whereas AGI's risks are far more abstract and unimaginable, which could lower the psychological barrier for both countries to develop and even employ it against each other.
“in the cold war we very clearly knew what nuclear weapons would do ... whereas here ... the risk would be a lot more abstract and unimaginable. So there would be less of a barrier”
The Cold War II scenario assumes both the US and China achieve AGI and are transformed by it while maintaining rough parity and competitive tension, with AGI accelerating but not fundamentally eliminating the competitive bilateral relationship
“cold war II is very much a scenario in which both the US and China achieve AGI are transformed by it but remain in sort of a competitive you know in a competitive environment of rough parody to each other where you know the invention of AGI has not necessar um has not necessarily uh you know in so far as it's impacted things has probably equalized the relationship more from the US perspective that may mean declined some of some advant advantages have declined and but it's a but you know you know it's a cold war it's not a hot war so it's bilaterally so there's a bilateral tension but a bilateral balance”
War requires deliberate decision-making by states rather than accidental escalation, though accidents and misinterpretation remain possible, so whether AGI causes war depends on whether states deliberately choose to fight over AGI advantages or other issues
“I do think you know there could be accidents AGI particularly in the cyber domain could be quite transformative for example in ways that are not stability enhancing that are stability deteriorating But I think that's sort of like a casebyase basis and you know whether major war breaks you know the relationship might be more negative or positive but whether major war breaks out I think it will come down to countries sort of decide like like major war will probably be a decision rather than an accident”
US government and AI company officials frequently cite China as a justification for not slowing AGI development or imposing regulations, arguing that unilateral restraint would hand technological advantage to China.
“when you when you ask or when people in in in the government or in the AI labs are asked about regulation often the answer uh that they give is that they can't really slow down because of China because that China would would um get ahead.”
The paper deliberately avoids assigning probabilities or timelines to the scenarios because the high uncertainty makes such estimates unreliable, and because the goal is to align readers to the magnitude of potential impact—which is enormous whether AGI arrives in 10 or 50 years—rather than to its timing.
“what we're really trying to align people to is the size of the the size of the potential impact not when it's going to show up”
Smith has personal doubts about whether AGI can be controlled as easily as nuclear weapons, suggesting he questions the validity of the 'AGI as nuclear weapons' analogy for proliferation control
“Just caveing the assumption that is controllable like nukes which I think over time I've I personally have doubts about the ability to control it in the way that nuclear weapons h have and we can talk more about that.”
There are roughly a hundred analogies people reach for when describing AI under uncertainty (nuclear weapons, electricity, fire, the internet), reflecting how people grasp for analogies amid uncertainty.
“there are about a hundred analogies for AI because we people reach for analogies and uncertainty”
As an analyst, it is always better to avoid getting too specific with dated predictions, because while precise estimates can be valuable to policymakers, surfacing the space of possibilities is more robust under high uncertainty.
“as an analyst, it's always better to not get too specific in predictions with with dates”
RAND has published extensively on nuclear war, arms control, and first-strike dynamics that inform contemporary thinking about AGI arms races and strategic instability
“technologically neither side could escape it for uh and this is an area that Rand has written a lot on. I personally am not but and if you search Rand nuclear war you will find plenty of of papers that sort of address this question”
The host perceives Cold War II, Authoritarian Advantage, and Wild Frontier as the three most likely scenarios, suggesting these have more realistic foundations than the other five scenarios presented
“I've actually as I was reading the report I even though you didn't assign probabilities to me I've had a few scenarios that sort of emerged as the ones that to me seemed like the most probable. Um, and the ones that I saw as the most likely were scenarios that you called War II, the authoritarian advantage, and the third was the wild frontier.”
Smith indicates that he has a view on which of the three scenarios the host identified (Cold War II, Authoritarian Advantage, Wild Frontier) is most likely but delays revealing it to the end of the conversation
“Well, I would agree that one of those three is the one I think is the most likely. just but I'll we can leave that as a mystery toward to the end and I'll uh which one of those three I think is the most likely.”