The Decoupling Scenarios

In Busan, South Korea, in the early months of 2026, American and Chinese negotiators sat across from each other and agreed to stop escalating — for one year. The Americans would hold their chip export controls at current levels. The Chinese would suspend their latest rare earth restrictions. Both sides would revisit the arrangement in twelve months.

It was the most significant moment of restraint in the AI infrastructure war. It was also, by every measure available, a pause — not a peace. Both delegations returned home and continued building parallel technology stacks. The agreement did not address cables, energy, minerals, or military AI. It addressed chips and rare earths. Two nodes among eight.

The question the Busan pause leaves unanswered is the question this chapter must confront: Where is all of this going?

Three futures are plausible. Each can be traced through the eight fronts this book has mapped. Each has evidence for and against it. And each carries consequences that extend far beyond the technology itself.

Scenario One: Managed Competition

This is the optimist's future — and the diplomat's.

In this scenario, the Busan pause is not a pause but a precedent. The one-year suspension extends into a longer framework. Norms emerge around AI infrastructure, modeled on the arms control agreements that — imperfectly, with violations and crises — prevented nuclear war for eighty years. The United Nations' 39-member Scientific Panel on AI, appointed in February 2026, produces assessments that both Washington and Beijing take seriously. The Global Dialogue on AI Governance convenes in Geneva in July 2026 and produces not just declarations but mechanisms — binding commitments, verification protocols, shared standards.

Trace this through the eight fronts:

Energy: Iran and the Gulf reach a fragile accommodation. Hormuz stays open, not because anyone trusts anyone, but because the economic incentive to keep oil flowing outweighs the strategic incentive to close the strait. China diversifies its energy imports — slowly, expensively — while the United States maintains its independence.

Chips: TSMC continues fabricating for both American and Chinese clients, with export controls that limit cutting-edge access but permit trade in mature nodes. Taiwan's silicon shield holds — not because anyone wants it to, but because the cost of breaking it is unthinkable.

Cables: An international framework for undersea cable protection emerges, perhaps modeled on the treaties that govern satellite communications. The Red Sea remains contested, but cable repair ships gain protected status.

Minerals: Chinese rare earth controls relax as Western alternatives come online — slowly, over ten to fifteen years. The processing monopoly erodes but does not collapse.

Capital: Gulf investment flows to both stacks, hedging. The Gulf states become the Switzerland of the AI wars — profiting from neutrality.

The evidence for this scenario is real but thin. Fifty-eight countries signed the Paris AI Action Summit declaration. The EU AI Act is operational, with penalties severe enough to concentrate corporate attention. The economic modeling is clear: the IMF estimates full decoupling would shave seven percent off global GDP — $7.4 trillion — and the US Chamber of Commerce projects annual costs of $738 billion. No rational actor wants this.

The evidence against it is thicker. The United States and the United Kingdom refused to sign the Paris declaration. The Trump administration rescinded the global AI Diffusion Rule. No binding enforcement mechanism exists for any international AI agreement. Both sides are building parallel stacks not because they want to decouple, but because they do not trust the other side not to weaponize interdependence. Trust, once broken, rebuilds on the timescale of generations, not negotiations.

Assessment: the weakest scenario. Not impossible — but requiring sustained diplomatic investment that current political dynamics do not support.

Scenario Two: Full Decoupling

This is the trajectory we are on.

Not a clean break — the supply chains are too entangled for that. But a progressive bifurcation into two parallel technology ecosystems, each optimized for independence from the other, each drawing allied and client states into its orbit.

The architecture is already visible. On the American side: NVIDIA GPUs fabricated by TSMC on ASML machines, running the CUDA software stack, deployed through AWS, Azure, and Google Cloud. On the Chinese side: Huawei Ascend chips fabricated by SMIC using deep ultraviolet lithography, running the CANN software stack, deployed through Alibaba Cloud, Tencent Cloud, and Baidu Cloud. DeepSeek releases models with first-class support for both ecosystems — because DeepSeek, like the hedging nations, understands that the future may require operating in two worlds simultaneously.

A third pole is forming. The Gulf states are building their own stack — from compute to applications — using capital as leverage. The EU is not building a frontier stack at all, but exercising regulatory sovereignty through the AI Act, with penalties of up to €35 million or seven percent of global turnover. Europe may not have the models, but it has the rules — and rules, in a bifurcated world, become a form of power.

Trace this through the fronts:

Energy: Decoupling accelerates the energy competition. China invests massively in nuclear and renewables to reduce its oil dependency — not because it wants to, but because it must. The Hormuz vulnerability becomes an existential motivator. The United States deepens its natural gas advantage and pushes SMR deployment. Energy independence becomes the prerequisite for technological independence.

Venezuela: In a decoupled world, Venezuelan oil becomes a strategic asset again. Whoever controls access to 304 billion barrels of proven reserves controls a pressure valve in the global energy market. Sanctions tighten or loosen based on which stack Venezuela's output feeds.

Ukraine: The gas war becomes permanent. Ukraine's neon industry — shattered by invasion, partially rebuilt, partially relocated — becomes a case study in supply chain fragility that neither side wants to repeat. Semiconductor gas production disperses globally, at higher cost.

Taiwan: The silicon shield cracks under pressure. Not military pressure — not yet — but economic and political pressure to choose. TSMC's overseas fabs in Arizona, Japan, and Germany become lifelines, not luxuries. The island's strategic position shifts from "too valuable to attack" to "too important not to secure" — a distinction that matters enormously to the people living there.

Minerals: China weaponizes its processing monopoly selectively — restricting exports to adversaries while maintaining supply to clients. The "choke" becomes a permanent feature of Chinese statecraft, not a one-time play. Western mining and processing projects accelerate, but the fifteen-year timeline means a decade of vulnerability.

Cables: Route diversification becomes existential. Arctic cables — Polar Connect, Far North Fiber — are built not for efficiency but for strategic redundancy. The Red Sea corridor is progressively abandoned by traffic that can afford the latency cost of alternatives. The cable map of 2035 looks nothing like the cable map of 2025.

Gulf: The Gulf states play both sides, building AI infrastructure with American technology and Chinese capital, or vice versa. The February 2026 missile strikes on Dubai — Iran's reminder that Gulf hubs exist within range — complicate the "stable hub" narrative but do not kill it. Capital flows where it can earn returns; the returns on AI infrastructure are too large to abandon.

Greenland: The Arctic play intensifies. Rare earth mines, data cables, and space infrastructure — the triple play — become the insurance policy for the American stack. Fifty-seven thousand Greenlanders find themselves at the center of a great-power competition they did not invite.

The economic costs are staggering. CEPR modeling projects welfare losses of one to eight percent for the Western bloc and eight to eleven percent for the Eastern bloc. Semiconductor decoupling alone costs an estimated $114 billion annually. Aviation: $38 to $51 billion a year. Medical devices: $23.6 billion a year.

But the Cold War proved that bifurcation is sustainable — for the wealthier party. The Soviet Union's technology stack was inferior but functional for four decades. China's stack is already more capable than the Soviet Union's ever was, and China's economy is vastly larger relative to America's than the Soviet economy was. This is not a rerun of the Cold War. It is a harder, more entangled, more expensive version — with the critical difference that the supply chains were never separate to begin with.

The countries in the middle — India, Indonesia, Brazil, Vietnam, Malaysia, Mexico — become "connector economies," profiting from the friction between blocs. They build in both stacks. They resist being forced to choose. They become the leaks in the wall, and both blocs tolerate them because the alternative — total separation — is too costly even for the superpowers.

Assessment: the current trajectory. Not full decoupling but selective bifurcation — parallel stacks with managed entanglements. A messy, expensive, durable equilibrium.

Scenario Three: Systemic Breakdown

This is the scenario that keeps strategists awake at three in the morning.

Not because it is likely. Because it is possible. And because the system is optimized for efficiency, not resilience — which means it has very little margin for error.

The scenario begins with simultaneous stress at multiple nodes. This is not hypothetical — it is what February 2026 looks like. US-Israel strikes on Iran threaten Hormuz. Houthi retaliation disrupts Red Sea cables. Rare earth controls tighten. Gulf investor confidence wavers. Taiwan Strait tensions elevate. Any one of these is manageable. The question is what happens when they converge.

Network theory provides the mathematics. In interdependent networks — systems where the failure of one component triggers failures in connected systems — a counterintuitive property emerges: the diversity that makes a single network robust makes an interconnected system fragile. Researchers have identified a critical fraction of failed nodes that triggers complete fragmentation — a phase transition, in the language of physics. Below the threshold, the system degrades gracefully. Above it, collapse.

The Italy blackout of 2003 is the textbook case. Power station failures caused internet node failures, which caused further power station failures — a feedback loop between interdependent networks that cascaded faster than operators could respond. It was resolved in hours. The AI supply chain, if similarly cascading, would not resolve in hours. A rare earth shortage that halts ASML shipments that delays TSMC production that constrains NVIDIA supply that slows data center deployment that reduces AI training capacity that degrades the military systems fighting the wars that started the cascade — this loop does not have an off switch.

Trace it through the fronts:

Energy + Cables: Hormuz closes and Red Sea cables are cut simultaneously. China loses both oil supply and data connectivity to the Middle East and Europe. Energy prices spike globally. Data center operating costs rise. Cloud services degrade.

Chips + Minerals: China retaliates with full rare earth embargo. TSMC production halts on advanced nodes within weeks as consumable stockpiles deplete. ASML cannot ship replacement parts. The $10 trillion estimated impact of a Taiwan conflict becomes a reality — not from invasion, but from supply chain asphyxiation.

Gulf + Arctic: Gulf AI hubs go offline as regional conflict escalates. Arctic infrastructure — the backup plan — is years from completion. There is no redundancy.

The Feedback Loop: The AI systems being used to manage the crisis — military AI, logistics AI, intelligence AI — degrade as the infrastructure that supports them degrades. The recursive loop turns vicious. The weapon depends on the supply chain that is collapsing.

The COVID semiconductor shortage was a preview — a single-domain disruption that took two years to resolve, in peacetime, with no active conflict at any major node. A multi-node cascading failure in the AI supply chain has no historical precedent. The models that exist — financial stress tests, infrastructure resilience assessments, military wargames — are designed for individual sectors, not for the interconnected system as a whole.

The evidence against this scenario is the system's demonstrated resilience. Red Sea cable cuts did not cause internet blackouts. The semiconductor shortage resolved. Markets adapt. Governments intervene. The incentives for stability are enormous.

But the incentives for stability were enormous in 1914, too. The interdependence of European economies was supposed to make war impossible. It did not. What interdependence created was not peace but a system where, once the cascade began, no single actor could stop it.

Assessment: low probability, catastrophic consequence. The scenario no one plans for — because planning for it requires acknowledging that the system's architects optimized for the wrong variable.

Three futures. One optimistic and unlikely. One costly and current. One catastrophic and conceivable.

The question that none of the scenarios answers — the question that no strategic model can answer — is whether there exists a fourth path. Not managed competition, which requires trust that does not exist. Not decoupling, which requires sacrifices no electorate has been asked to accept. Not breakdown, which requires a failure of imagination that the existence of this book is designed to prevent.

Something else. Something that begins with seeing the system whole — not as eight separate fronts, not as a bilateral competition, not as a technological race, but as a single interconnected field whose health depends on the health of every node, including the nodes inhabited by people who never asked to be part of this war.

To see it that way requires a different lens.