“The future is already here — it’s just not evenly distributed.”
— William Gibson, Novelist, 1993
Gibson’s observation about distribution is the key to understanding every long-run AI scenario. The question is never whether advanced AI will arrive — it will. The question is who governs it, who owns it, who bears the cost of its infrastructure, and whether the political economy produces equitable distribution of its surplus or its concentration. The four scenarios below are not equally probable. They are the four destination states that current trajectories point toward, depending on which governance choices are made — or not made — in the next decade.
Scenario Architecture
What Leads Where: The Decision Tree
The four scenarios are not independent futures — they are outputs of a branching decision tree whose critical nodes are governance choices, energy policy, and competitive dynamics. The flowchart below maps the pathways from current conditions to each destination state.
AI Scenario Pathway Flowchart — From Present Conditions to Long-Run Outcomes
All scenario probabilities are Fenrir Research analytical estimates. The branching structure reflects key governance and policy variables, not a deterministic model.
Section 02
The Four Scenarios in Detail
Scenario A · 15–20% Probability
The Productivity Utopia
Conditions Required: Effective international AI governance · Rapid grid decarbonisation · Antitrust action preventing monopoly rents · Universal access frameworks
In this scenario, AI’s productivity dividend is real and broadly distributed. AGI-adjacent systems compress the innovation cycle across energy, medicine, and materials science — driving economic growth rates that have no recent historical precedent. Former OpenAI researcher Leopold Aschenbrenner’s modelled thesis of 30%+ annual growth in the superintelligence era represents the upper bound of this scenario’s economic impact.
Human labour is not eliminated but reallocated: cognitive tasks are augmented rather than replaced, with the freed capacity redirected toward care work, creative industries, political deliberation, and leisure. Universal Basic Income or equivalent redistribution mechanisms ensure productivity gains flow broadly rather than concentrating in capital. Clean energy transition — driven partly by AI-accelerated materials science for batteries and solar — means the AI infrastructure buildout does not compound the climate problem.
The reason this scenario is assigned low probability is not that it is technically impossible. It is that it requires a level of international coordination, political will for redistribution, and antitrust enforcement against the most powerful companies in modern economic history that has no clear precedent. The analogy is not the internet’s democratisation of information — it is the management of nuclear technology, which required decades of treaty-building and still has not fully succeeded.
Scenario B · 25–30% Probability
The Resource Depletion Scenario
Conditions Required: Governance partial · AI deployment accelerates but energy decarbonisation lags · Water-stressed regions bear infrastructure costs · Critical mineral supply chains tighten
In this scenario, AI deployment continues at current pace but the energy and resource infrastructure required to support it conflicts with climate and resource sustainability objectives. Data center electricity demand, projected to more than double by 2030 from its current 4% share of U.S. electricity, is met primarily by gas-fired generation — adding to carbon lock-in at precisely the moment when the energy system needs to be decarbonising fastest.
Water scarcity becomes a structural constraint rather than a manageable operational cost. In water-stressed regions — the American Southwest, parts of India, the Middle East — data center water demand competes directly with agricultural and human consumption needs. The political economy of this conflict is not resolved smoothly: communities host the infrastructure costs while the economic benefits accrue elsewhere.
Critical mineral supply chains — cobalt, lithium, neodymium, and the gallium and germanium that semiconductor supply chains depend on — face concentration risk as demand scales. The Democratic Republic of Congo’s cobalt dominance and China’s control over rare earth processing represent systemic vulnerabilities that an accelerating AI buildout intensifies. This scenario ends not with civilisational collapse but with constrained AI deployment, higher input costs, and a significant exacerbation of the climate transition challenge.
Scenario C · 35–40% Probability — Most Probable Current Trajectory
The Corporate Oligarchy
Conditions Required: Governance absent or captured · No effective antitrust · AI moats deepen · Labour displacement unaddressed · Capital concentration compounds
This is the scenario that current trajectories most clearly point toward, and it is worth stating its logic plainly. A small number of corporations — perhaps five to ten globally — develop AI systems that are sufficiently capable and deeply integrated into economic infrastructure that switching costs and network effects create persistent monopoly positions. Those positions generate supernormal returns that fund further AI capability development, deepening the moat. Antitrust frameworks designed for the industrial economy cannot dismantle software-and-data advantages at current enforcement speed.
The distributional consequences are severe. AI-driven productivity gains accrue primarily to shareholders of the companies owning AI systems and to the high-skill workers whose labour is complemented rather than substituted. Workers in cognitive roles that AI substitutes — junior analysts, paralegals, customer service, entry-level coders — face structural unemployment in sectors where no comparable transition roles exist. The “jobs will be created” argument of prior automation waves may be partially true in aggregate but does not address the transition cost for the specific workers displaced.
Politically, this scenario produces the conditions that historically precede significant social instability: concentrated wealth, visible technological unemployment, and the perception — grounded in reality — that the gains of a major economic transformation are not being shared. The policy responses — whether redistributive taxation, UBI experiments, or more disruptive political outcomes — become the defining political economy question of the 2030s.
Why This Is the Most Probable Scenario
The Corporate Oligarchy scenario does not require any unusual conditions — it is the default output of current trajectories absent deliberate intervention. Regulatory frameworks are lagging. Antitrust enforcement is slow. Capital is concentrating. The political coalitions necessary for redistribution frameworks have not formed. This is not a pessimistic reading; it is the base case, which makes intervention — if it is to happen — urgent.
Scenario D · 20–25% Probability
The Geopolitical Fracture
Conditions Required: US-China AI decoupling deepens · No international treaty framework · AI arms race in autonomous weapons · Developing world as battlefield for AI influence
The US-China semiconductor and AI decoupling that began with export controls on advanced chips in 2022 has, by 2025, produced two increasingly incompatible AI ecosystems: one anchored in US-designed chips and US-based foundation models, the other anchored in Chinese semiconductor alternatives and Chinese-developed foundation models. If this trajectory continues, the result is a digital iron curtain with significant implications for the interoperability of global financial systems, scientific collaboration, and international communication infrastructure.
Autonomous weapons systems represent the most dangerous dimension of this scenario. The absence of any binding international treaty framework governing lethal AI systems — while all major military powers pursue capability in this domain — creates conditions analogous to the pre-arms-control environment of early nuclear technology, but with a faster development cycle and lower barriers to entry. A single escalation involving AI-enabled autonomous systems in a contested geography (Taiwan Strait, South China Sea, Eastern Europe) could trigger escalation dynamics that existing deterrence frameworks are not designed to manage.
The developing world dimension of this scenario — smaller nations becoming sites of AI infrastructure investment and data extraction from both US and Chinese tech ecosystems, without meaningful sovereignty over the AI systems shaping their information environments and economic decisions — represents a new form of technological dependency that is not captured in existing international development frameworks. AI colonialism, where a nation’s language, culture, and data train systems that are then sold back at monopoly prices with no local governance, is already emerging as a structural pattern.
Section 03 — New Scenario Layer
Resource Wars in the AI Age: The Fifth Scenario No One Is Pricing
The four scenarios in Part IV’s framework address governance, energy, corporate concentration, and geopolitical decoupling. There is a fifth dynamic that sits underneath all of them and has not received adequate analytical attention in most AI-focused research: the extent to which AI’s physical resource requirements are already reshaping state behaviour, military posture, and the calculus of territorial control. This is not a future risk. It is a present one — and the pattern of recent US foreign policy activity provides the clearest evidence.
The Observation That Requires Explanation
In 2024–25, the United States has maintained or escalated pressure on Venezuela (oil), expressed explicit interest in acquiring Greenland (rare earths, Arctic resources, strategic depth), re-tightened sanctions on Iran (oil, regional influence), and maintained Cuba policy partly through the lens of Chinese infrastructure presence in Havana. These are not isolated diplomatic episodes. They form a coherent pattern: resource-securing and competitor-excluding behaviour by a state that has concluded, at the strategic level, that the physical inputs to AI dominance are geopolitical prizes worth contesting.
Why AI Creates Resource Imperatives for States
The connection between AI capability and physical resource control runs through three pathways. First, compute requires minerals that are geographically concentrated and partially controlled by adversary states — gallium, germanium, rare earths, cobalt. A state that controls or secures alternative supply chains for these materials has structural advantage in the AI race that persists regardless of software advances. Second, AI infrastructure requires power at unprecedented scale — and the states with the largest clean energy endowments (hydroelectric, geothermal, wind, solar) will host AI infrastructure at lower cost and with lower climate liability than those without. Third, the economic surplus generated by AI dominance is large enough to fund military and geopolitical projection at a scale that makes the resource competition self-reinforcing.
The US-China dynamic makes this explicit. China’s October 2023 export restrictions on gallium and germanium were not merely a trade retaliation to US chip controls — they were a demonstration of the resource leverage that China holds over the semiconductor supply chain. China controls approximately 80% of global gallium production and 60% of global germanium refining. A sustained supply restriction would not halt US AI development overnight, but it would impose significant cost increases and sourcing delays on the chip manufacturing base that US AI depends on. China’s rare earth processing dominance (roughly 85% of global capacity) amplifies this leverage. When Beijing discusses “resource weapons,” it is not being metaphorical.
| Resource / Geography |
AI Relevance |
Current Control |
US Strategic Interest |
Recent Activity |
| Greenland — Rare Earths, Arctic |
Largest known rare earth deposits outside China; Arctic as strategic infrastructure corridor |
Danish sovereignty; Greenlandic autonomy |
Explicit acquisition interest stated by Trump administration (2019, 2025) |
Repeated US overtures; Danish diplomatic tensions; Chinese mining consortium previously rejected |
| Venezuela — Oil, Coltan |
Coltan (tantalum/niobium) used in capacitors for electronic devices; oil revenues fund rival AI infrastructure |
Maduro government; Chinese and Russian investment presence |
Sanctions maintained; competitor exclusion; potential future access |
OFAC sanctions cycles; oil sector waiver experimentation; coup monitoring |
| DRC — Cobalt, Coltan |
~70% of global cobalt (EV batteries, UPS systems); significant coltan deposits |
Fractured; Chinese companies hold majority of major cobalt mines |
Medium; Lobito Corridor rail investment signals |
Lobito Atlantic Railway (US/EU-backed); competing with Chinese Belt & Road mining infrastructure |
| Iran — Oil, Regional Influence |
Indirect: oil revenues fund AI-adjacent military capability development; Strait of Hormuz choke point for Gulf AI infrastructure energy supply |
Sanctioned; China as primary oil buyer |
Prevent Chinese-Iranian energy partnership from funding adversary capability |
Renewed maximum pressure (2025); sanctions on Chinese firms buying Iranian oil |
| Cuba — Chinese Intelligence Presence |
China operating signals intelligence station in Cuba (reported 2023); proximity to US AI infrastructure in Florida/Southeast |
Cuban sovereignty; Chinese lease agreements |
Exclusion of Chinese intelligence infrastructure from Western Hemisphere |
US pressure on Cuba to terminate Chinese facility agreements; continued embargo |
| Ukraine — Rare Earths, Titanium |
Significant deposits of lithium, titanium, neon gas (used in chip fab lasers), rare earths in eastern regions |
Contested; Russian-occupied territories contain major deposits |
Minerals deal framework agreed between US and Ukraine (2025); resource access explicitly in negotiation |
US-Ukraine Critical Minerals Agreement (Mar 2025); Russian consolidation of mineral-rich Donbas regions |
| Canada — Critical Minerals |
Major lithium, cobalt, nickel, and rare earth deposits; politically stable, allied supply |
Canadian sovereignty; increasing US pressure |
Trump tariff pressure partly a resource-access negotiation posture |
Tariff disputes (2025); US pressure for deeper resource-sharing frameworks; Canada pursuing independent processing capability |
The Ukraine Minerals Dimension
The clearest example of AI resource logic entering active geopolitical negotiation is Ukraine. The minerals agreement between the United States and Ukraine, signed in principle in early 2025 after intense negotiation, explicitly linked US security guarantees and reconstruction support to US access to Ukrainian critical minerals — specifically the titanium, lithium, and neon gas deposits that are strategically significant for semiconductor manufacturing. Ukraine’s geological survey estimates it holds approximately 20% of global titanium reserves, significant lithium brine deposits in the Donetsk region (currently occupied), and is a major source of neon gas used in the excimer lasers that ASML lithography equipment depends on.
This is not coincidental. It is a signal that the US has concluded — at the strategic planning level — that the critical mineral supply chain for AI and semiconductor manufacturing is a national security priority equivalent to oil supply security in the 20th century. The minerals deal with Ukraine is the first overt example of AI resource logic shaping US foreign and security policy in a live conflict context.
Greenland: The Starkest Resource Case
The Trump administration’s publicly stated interest in acquiring Greenland — expressed in 2019 and returned to with greater intensity in 2025, including suggestions that military options were not off the table — is best understood through this resource lens rather than as idiosyncratic behaviour. Greenland holds one of the world’s largest undeveloped rare earth deposits (Kvanefjeld/Kuannersuit), significant oil and gas resources, and occupies a strategic position in Arctic routes that are becoming commercially and militarily significant as ice cover declines. It also represents an opportunity to deny Chinese mining access — multiple Chinese mining consortium bids for Greenlandic mineral projects have been blocked on national security grounds.
From a conventional diplomacy perspective, the suggestion of acquiring an autonomous territory of a NATO ally is extraordinary. From an AI resource strategy perspective, it is consistent: Greenland’s rare earth deposits, if developed under US-aligned governance, would represent a material reduction in Chinese leverage over the semiconductor supply chain. The willingness to absorb the diplomatic cost of making this interest explicit — even if the acquisition itself never occurs — signals how seriously the resource competition is being taken at the policy level.
From Chip War to Resource War: The Escalation Pathway
The current phase of competition — chip export controls, mineral export restrictions, intelligence infrastructure exclusion — is best understood as the opening moves of a resource competition whose ultimate stakes are AI dominance. The escalation pathway is not hard to trace: if AI capability compounds with compute access, and compute access depends on mineral supply chains, and mineral supply chains are concentrated in politically contested geographies, then the logic of securing those supply chains through political, economic, or — in the limit — military means is structurally analogous to 20th century oil resource competition.
The 2003 Iraq War is instructive not as a direct precedent but as a demonstration of how states behave when they conclude that a critical resource is controlled by a hostile actor and that control is existentially important. The calculation failed in execution, but it was rational in its premise. A state that concludes that rare earth and semiconductor mineral supply chains are the oil of the AI economy will eventually develop a foreign policy that reflects that conclusion — whether through commercial competition, political pressure, sanctions, or ultimately something harder. The pattern visible in 2024–25 US foreign policy suggests that conclusion has already been reached at the strategic planning level. The overt component of the response is early stage. The covert component — intelligence operations, grey-zone activity around mining infrastructure, competitive investment in alternative supply chains — is likely further advanced than is publicly visible.
Portfolio Implication — Resource War Premium
The resource competition dimension adds a risk premium to any AI investment thesis that depends on uninterrupted access to Chinese-controlled mineral supply chains. Gallium, germanium, and rare earth processing dependencies are not priced into semiconductor equity multiples at current levels. The investment case for non-Chinese rare earth producers, North American lithium developers, and the Lobito Corridor infrastructure investments (which create alternative DRC cobalt routing outside China’s logistics control) is partly a commodities call and partly a geopolitical risk hedge — and it is one of the more under-owned trades in the current AI cycle.
Section 04 — Structural Question
Does AI Reverse Multipolarity? The Case for a New Unipolar Moment
The post-Cold War international order has been characterised — with increasing accuracy since roughly 2008 — as multipolar: the United States as the leading but no longer unchallenged power, with China, the EU, Russia, India, and regional powers each asserting spheres of influence that constrain American primacy. The question that AI raises, and that most geopolitical analysis has not yet seriously engaged with, is whether AI capability sufficient to produce an overwhelming and durable advantage in economic output, military capability, and intelligence processing would structurally reverse multipolarity — effectively reproducing a unipolar world, but with AI rather than nuclear weapons and dollar dominance as the foundation of power.
The Argument for AI-Driven Unipolarity
The argument rests on three claims. First, AI capability advantages compound: a state with a significant lead in frontier AI at T₀ will have faster economic growth, better military capability, and more effective intelligence at T₁ — and those advantages fund further AI investment, widening the lead at T₂. Unlike conventional military hardware, which depreciates and requires maintenance, AI models improve in deployment and generate the data that trains successor models. The advantage is self-reinforcing in a way that most prior sources of national power are not.
Second, the inputs to AI are heavily concentrated geographically and institutionally. Frontier AI development currently occurs at perhaps five to eight organisations globally, all of them in the United States or closely allied with US institutions. The compute substrate — Nvidia GPUs, TSMC-fabricated chips, hyperscaler cloud infrastructure — is predominantly US-designed or US-allied. The training data advantages accrue to organisations with access to the largest English-language internet corpora, which are US-domiciled. China is the credible second, but the gap between US and Chinese frontier capability — while smaller than three years ago, partly due to DeepSeek’s efficiency innovations — remains material. No other actor is currently in serious contention.
Third, AI-enabled economic productivity, if it compounds at rates that some researchers project, could produce US economic output growth that leaves the rest of the world so far behind that the structural basis of multipolarity — rough economic parity among major powers — is simply eroded away. A United States growing at 10%+ annually because of AI-driven productivity while China grows at 4–5% and the EU at 2–3% would, within a decade or two, have an economic base so large relative to competitors that conventional balance-of-power dynamics cease to operate.
Hypothetical GDP Trajectory Under AI Dominance — Indexed (US=100 in 2026), 2026–2040
Illustrative scenario only. GDP indexed to US = 100 in 2026. US growth rate assumes AI-driven productivity uplift approaching Goldman Sachs upper scenario. Actual trajectories will depend on diffusion, governance, and competitor AI adoption. Not a forecast.
The Counterarguments — Why Unipolarity May Not Materialise
The unipolarity thesis, while analytically coherent, faces several serious counterarguments that Fenrir believes are strong enough to prevent a clean outcome.
Diffusion is not containable. The history of transformative technology is that advantages erode faster than incumbents expect, because knowledge diffuses and because the technology itself lowers the cost of subsequent adoption. DeepSeek-R1’s demonstration that frontier-capable models can be trained at dramatically lower compute cost than previously assumed is a concrete example: Chinese researchers, denied access to the best hardware, innovated around the constraint and produced architectures that reduce the compute advantage of US AI labs. This pattern — disadvantaged actors innovating more efficiently out of necessity — is well-documented in technology history. It is not clear that the US lead in frontier AI is as durable as the unipolarity thesis requires.
Multipolarity has institutional momentum. The international institutions, alliances, trade frameworks, and legal norms built around a multipolar world — the UN Security Council structure, WTO, the IMF voting framework, regional security architectures — do not simply dissolve because one power’s GDP grows faster than others. They create friction, delay, and political costs for unilateral action that constrain even a dramatically more powerful United States. The US already has the world’s largest economy and most powerful military, and it has not achieved unipolar dominance. The question is whether AI advantage is qualitatively different — and that depends on how quickly it translates into deployable geopolitical power versus GDP growth that takes years to convert into military and institutional leverage.
Coalition against dominance is the most reliable dynamic in international relations. If it becomes apparent that AI is producing a durable and accelerating US power advantage, the political incentive for other major powers — including current US allies who are not themselves on the frontier — to form countervailing coalitions is strong. This is the oldest mechanism in the balance-of-power tradition, and it has never been more dormant than for a generation at a time. An AI-powered US that overreaches on resource control, technology exclusion, and economic coercion would likely accelerate rather than prevent coalition formation against it.
The honest assessment: AI is more likely to produce a new phase of contested unipolarity — in which the United States is structurally the most powerful actor but does not achieve the uncontested dominance that the term “unipolar” implies — than to resolve the current multipolar tension definitively in either direction. The resource competition, the chip war, and the geopolitical manoeuvrings are not the prelude to a stable new order; they are the symptoms of a transition period whose outcome is genuinely uncertain.
Analytical Judgement — AI and Geopolitical Structure
The scenario that Fenrir assigns highest probability is not AI-driven unipolarity but something more unstable: a period in which the United States has a significant but not decisive AI advantage, China has a large domestic AI capability that is partially but not fully constrained, and the rest of the world is effectively dependent on one or both of these two ecosystems for AI infrastructure. This is not multipolarity in the conventional sense — the AI-mediated dependencies are too asymmetric. But it is not unipolarity either — the resource competition, coalition dynamics, and diffusion of capability prevent clean dominance. Call it contested primacy: a structural configuration more dangerous than either stable unipolarity or stable multipolarity, because it combines strong incentives for resource competition with insufficient dominance to deter challengers. It is the configuration most likely to produce the resource conflict scenarios described in Section 03 of this post.
Section 05
The Governance Variable: Why the Next Five Years Are Critical
The flowchart’s most important node is not the start state (AI rapid advancement) or the end states (the four scenarios) but the decision nodes in between. Technology paths exhibit strong path dependency: once infrastructure is built at scale, once regulatory frameworks are established, once competitive positions are locked in, reversing direction is enormously costly. The window for governance intervention that shifts the trajectory from Scenario C or D toward Scenario A is finite and closing.
| Governance Lever |
Current Status |
What Would Shift the Scenario |
Timeline |
| Compute Governance |
US chip export controls; no multilateral framework |
International compute registry; access guarantees for developing nations |
Urgent: 2026–28 |
| Antitrust / Competition |
Existing cases (Google, Meta); no AI-specific framework |
Data portability mandates; mandatory API access; structural separation |
Urgent: 2026–29 |
| Labour Redistribution |
No major economy has enacted AI transition support at scale |
UBI pilots; retraining mandates; AI productivity levy |
Near-term: 2027–32 |
| Autonomous Weapons Treaty |
UN CCW talks stalled; voluntary moratoriums only |
Binding treaty with verification mechanism (analogous to NPT) |
Urgent: 2026–28 |
| Energy & Carbon Policy |
Data center exemptions from renewable mandates in some jurisdictions |
Mandatory renewable matching; carbon price on AI inference |
Near-term: 2027–30 |
| Training Data Rights |
Litigation underway; no settled framework in any jurisdiction |
Mandatory licensing scheme; creator compensation fund |
Medium-term: 2028–33 |
The EU AI Act represents the most developed governance framework currently in force, but its limitations are instructive. It addresses risk tiers for deployed applications but is weaker on foundation model obligations. Its enforcement depends on national market surveillance authorities whose capacity varies enormously across EU member states. It applies within the EU but not to AI systems deployed globally that affect EU citizens. And it was negotiated before the current generation of autonomous agent systems existed — meaning its risk classification framework is already partially obsolete.
The Nuclear Analogy
The most instructive historical precedent for AI governance is not the internet (which developed largely without governance infrastructure and whose negative externalities — disinformation, surveillance capitalism, platform monopoly — are now clear) but nuclear technology, which developed with intensive governance investment from the start. The Baruch Plan, the NPT, the IAEA inspection regime, and the CTBT were all costly and imperfect — but they created a framework that prevented the worst-case outcome. AI does not yet have an equivalent. Building one requires treating AI as the civilisationally significant technology that it is — not as a commercial product whose governance can be deferred until the technology is entrenched.
Leave a Reply