Part of the AI, Future and War series. Analysis and hypothetical examples are identified in the text.
Scenarios are a way to question assumptions
No article can establish how AI will shape conflict by 2035 with certainty. Technological performance, political choices and institutional behaviour interact in ways that resist a single confident forecast. Scenarios are useful because they expose different conditions under which a strategy might fail.
The three scenarios below are original analytical constructions, not predictions, probability estimates or descriptions of current events. UNIDIR's discussion of AI-related security risks and the IEA's work on AI's energy dependence provide background for considering both institutional and physical constraints. UNIDIR: AI and international security — risks and confidence-building IEA: Energy and AI, 2025
Scenario one: capability grows with stronger restraint
In this scenario, AI becomes more useful for organising information, but institutions also strengthen testing, accountability and communication. Important decisions retain deliberate review. Suppliers compete partly on evidence quality, controllability and the ability to explain limitations.
The implication for business is not that geopolitical risk disappears. It is that the use of AI becomes more legible. Organisations could invest in interoperable records, evidence-preserving workflows and staff who can challenge automated analysis. Indicators to watch would include demonstrable incident reporting, independent evaluation and sustained cooperation, rather than reassuring announcements alone.
Scenario two: speed outruns institutional control
Here, competitive pressure rewards fast adoption and rapid decision-making. Organisations use similar tools without understanding shared dependencies. Conflicting information circulates quickly, while staff have less time to verify its origin. A technically advanced information environment becomes vulnerable to coordinated mistakes.
A resilient business response would maintain independent sources, clear authority to pause consequential actions and procedures for verifying urgent requests. Tabletop exercises could test how the organisation responds when several trusted services repeat the same unverified claim. The scenario does not assume that AI inevitably causes conflict; it examines what happens when institutions fail to adapt.
Scenario three: fragmentation limits access
In this scenario, infrastructure constraints and geopolitical divisions make AI access less uniform. Service availability, data movement and supplier relationships become harder to manage across regions. Some organisations retain advanced tools while others operate with restricted or intermittent capability.
The practical response would emphasise portability, validated alternatives and a realistic minimum service. A business should know which tasks can be deferred, which require local capability and which depend on a supplier it cannot readily replace. This scenario is about operational choices under constraint, not an assertion about future export-control law or a named government's intentions.
Use the scenarios together
A strategy should be tested against all three futures. A plan that works only if compute remains cheap, suppliers remain accessible and information remains trustworthy is fragile. Equally, preparing only for the most adverse future may impose unnecessary cost and reduce useful innovation.
Review indicators periodically and record what would change the organisation's view. Avoid assigning numerical probabilities unless there is a defensible method and sufficient evidence. The aim is to identify decisions that remain useful across several plausible conditions, such as strong records, competent review and rehearsed continuity.
The leadership question
The important question for 2035 is not simply how intelligent AI will become. It is whether institutions can use greater capability without losing the ability to understand, challenge and recover from consequential mistakes. Leaders can begin answering that question now through the design of their decision processes, even while the technology's long-term trajectory remains uncertain.