The Map

In a glass-walled office on Lime Street in the City of London, a team of actuaries at Lloyd's of London is trying to price the end of the world.

Not the biblical version. The version where three things go wrong at the same time — where the Strait of Hormuz closes on the same day a cable is severed in the Red Sea and a Chinese export control catches a shipment of gallium destined for a chip packaging facility in Arizona. Lloyd's calls these "Realistic Disaster Scenarios," and every syndicate in the market is required to model them for capital adequacy. In 2025, they published one: a cyberattack on financial payments infrastructure causing $3.5 trillion in global losses. The scenario assumed a single point of failure.

No published Lloyd's model addresses what happens when multiple points fail simultaneously. This is not because the scenario is implausible. It is because it is happening right now, in February 2026, and the models cannot keep up with reality.

For eleven chapters, this book has traced the AI supply chain through eight separate conflict zones — each presented, by necessity, as a distinct front with its own geography, its own combatants, its own stakes. Iran and the energy that powers data centers. Ukraine and the gases that etch chips. The Red Sea and the cables that carry data. Taiwan and the fabrication plants that make the silicon. China and the minerals that go into everything. The Gulf states and the capital that funds the buildout. Greenland and the backup plan for all of the above. Venezuela and the locked-away oil reserves that could reshape global energy markets.

Presented individually, each front is alarming. Presented together, they reveal a single system with extraordinary concentration at every critical node, where disruption at one point cascades through all the others in ways that no existing model fully captures.

This chapter is the map. Not a metaphorical one. A literal topology of dependencies that, once seen, cannot be unseen.

So instead of listing the fronts again, walk through what is actually happening — right now, on this day — and watch the cascade.

It is February 28, 2026. The strikes on Iran began hours ago. The Strait of Hormuz is shutting down. Twenty million barrels of oil per day — one-fifth of global consumption — stops moving. Oil futures, which closed Friday at seventy-two dollars, will open on Monday into a market where analysts project prices between eighty and one hundred thirty dollars per barrel, depending on how long the Strait stays closed.

That price shock does not stay in the energy market. It propagates.

First, it hits shipping. The same Bab el-Mandeb Strait that Houthi attacks have disrupted for two years is now in the blast radius of Iranian retaliation. Insurance underwriters who already raised Red Sea war-risk premiums to one to two percent of vessel value suspend coverage entirely. The cable-laying ships working on Meta's 2Africa — already delayed by Houthi conflict — pull out of the Red Sea. The repair ships that would fix the next cable cut cannot enter the war zone. Fifteen submarine cables carrying seventeen percent of global internet traffic sit unprotected in a strait where missiles are now flying.

Second, it hits data. If even two or three of those cables are severed — by retaliatory strikes, by a sinking tanker's anchor, by Iranian Ghadir-class submarines deployed to the shallow eastern channel — twenty-five percent of Asia-Europe data traffic degrades. Microsoft Azure's Middle East cloud services, already strained by previous cable cuts, face latency surges. AI inference services in the Gulf, South Asia, and East Africa slow. The distributed training runs that some AI labs have begun operating across continents become impractical.

Third, it hits the Gulf. Iranian ballistic missiles are striking facilities in Bahrain, Kuwait, Qatar, and the UAE. The $2.5 trillion in Gulf AI investment — the Stargate campus, the HUMAIN initiative, the sovereign wealth plays — sits in the blast radius. Investor confidence, already shaken, cracks. The capital that funds the data centers that order the chips that TSMC fabricates in Taiwan is suddenly uncertain.

Fourth, it hits chips. Not directly — Taiwan is thousands of kilometers from the Persian Gulf. But TSMC's advanced fabrication depends on Chinese-sourced consumables for thirty percent of its manufacturing inputs. If Beijing, watching the US-Israel strikes on its primary oil supplier, decides to escalate its existing rare earth and mineral export controls — as it has done before, in October 2025, with extraterritorial restrictions on any product containing more than 0.1 percent Chinese-sourced materials — the ripple reaches Hsinchu. Advanced packaging, already the industry's primary bottleneck, tightens further. High-bandwidth memory, sold out through 2026, becomes unsourceable.

Fifth, it hits the broader economy. Oil above one hundred dollars triggers what the Dallas Federal Reserve models as a 1.3-percentage-point increase in annualized inflation. At one hundred twenty, the effect compounds further. A prolonged Hormuz closure is described in economic literature as a "guaranteed global recession trigger." In a recession, the hundred-billion-dollar training runs get delayed. The data center construction slows. The engineers who build the infrastructure demand higher wages because their groceries cost more. The money that funds AI dries up — not because the electrons stopped flowing but because the economy that produces the money contracted.

None of this is a prediction. Every link in that chain is documented in the preceding chapters. The cascade is not hypothetical — it is the architecture, operating under stress, on a day when the stress is real.

No published model combines this. Individual scenario models exist in isolation — the Rhodium Group's Taiwan analysis, Oxford Economics' Hormuz projections, CEPR's Red Sea trade models. Each prices a single disruption. Nobody has stress-tested the system for what is actually happening: multiple nodes failing or threatened simultaneously.

This is itself a finding. The most consequential infrastructure system in human history is not being stress-tested for its actual risk profile.

The historical precedents are not encouraging. In 2008, concentrated risk in financial instruments — collateralized debt obligations, credit default swaps — produced a systemic collapse when what appeared to be isolated failures turned out to be interconnected. The "bullwhip effect" amplified demand contraction as disruption propagated upstream through supplier tiers. Research published in Nature Scientific Reports found that only 0.035 percent of companies carry "extraordinarily high economic systemic risk," but their failure impacts twenty-three percent of national economic production. TSMC, SK Hynix, ASML, and NVIDIA occupy that same structural position in the technology domain.

COVID-19 demonstrated the same pattern in physical supply chains. Just-in-time systems that had "reduced buffers and made companies very reliant on quick and dependable cross-border flows" buckled when a single variable — pandemic lockdowns — disrupted production. The AI supply chain has inherited that same architecture: optimized for efficiency, catastrophically vulnerable to disruption.

The insurance industry understands this, even if it cannot fully price it. Swiss Re warned in 2025 that the AI boom "reshapes risk, but leaves insurers exposed." Specialized AI infrastructure risk facilities have begun to emerge, backed by major reinsurers. But even the largest — measured in hundreds of millions — are gestures against potential losses measured in trillions.

The system's defenders argue, correctly, that it has proven more resilient than feared. The neon gas disruption from Ukraine was absorbed within months as recycling technology was deployed and alternative sources developed. COVID chip shortages resolved within two years. Markets adapt.

But each of these was a single-node failure with time to respond. The neon supply recovered because the chip fabrication plants were still running, the cables were still carrying data, the energy was still flowing, the minerals were still being processed. Remove two or three of those conditions simultaneously, and the adaptive capacity that absorbed the neon shock does not exist.

AI efficiency improvements offer another counterargument. DeepSeek's R1 model demonstrated comparable performance to larger models at a fraction of the compute cost, suggesting the system might become less resource-intensive over time. But the history of technology suggests otherwise. Jevons paradox — the observation that efficiency gains are consumed by increased demand — has held for coal, oil, electricity, and computing. DeepSeek itself triggered a surge in Chinese AI chip demand, not a reduction.

The map is not a prediction. It is a description of what exists — a system of extreme concentration at every critical node, with interdependencies that multiply risk rather than distributing it, being operated by nations and corporations whose stress tests model individual failures in a world where failures come in clusters.

McKinsey put it plainly in 2025: "Most companies understand their direct vendors, but they lack visibility into the vendor networks behind them, where vulnerabilities can quickly cascade into real outages."

The map says: the cascades are not hypothetical. They are the architecture.