A picture of an Amazon Web Services data center in Manassas, Virginia, taken from above, with its rows of identical gray buildings sitting on flat land like a small city that exists only to think, has been making the rounds in financial and tech circles lately.
The incredibly long, delicate, geographically concentrated chain of materials, energy sources, and shipping routes that must all work flawlessly and simultaneously for that building to do anything at all is what makes the image striking, not what it depicts. A narrow strait in the Persian Gulf, a gas terminal in Qatar, or a chip factory in Taiwan are just a few examples of how one thread can alter the entire scene. In 2026, multiple threads are being pulled simultaneously.
| Topic | AI Boom Vulnerability & Multidimensional Economic Crisis (2026) |
|---|---|
| Core Thesis | Global economy dangerously concentrated in AI investments; supply chain exposed to Middle East instability |
| Key Authors | Matteo Wong & Charlie Warzel, The Atlantic (March 26, 2026) |
| AI Investment Scale | Trillions of dollars globally; in late 2025, virtually all U.S. economic growth came from AI investment |
| Critical Supply Chain Nodes | Advanced chips: primarily two South Korean companies + one Taiwanese firm |
| Energy Dependency | South Korea & Taiwan source majority of crude oil and LNG from Persian Gulf |
| Key Chokepoint | Strait of Hormuz — described as “critical to basically every aspect of the global economy” |
| Financial Risk | AI firms carrying historic debt loads; banks and private credit deeply interlinked |
| Token Economics Risk | Cost per AI token falling rapidly — described by analysts as “a death spiral to zero” |
| Historical Parallel | Compared structurally to 2008 financial crisis dynamics |
| Key Voices | Sam Winter-Levy (Carnegie Endowment); Brad Lipton (Roosevelt Institute) |
| Reference | The Atlantic — Welcome to a Multidimensional Economic Disaster |
Almost all of the US economy’s growth in the last few months of 2025 came from investments in AI. Let’s take a moment to process that sentence. Not most growth. A plurality, not. It’s all functional. Over the course of about three years, the tech sector was able to establish itself as the economy’s main engine rather than just a sector.
Depending on how closely you look at what’s underneath, this accomplishment is both impressive and unsettling. Data centers, chips, power infrastructure, cooling systems, and the businesses developing software to run on top of it all received trillions of dollars. It was a historic investment. It was also financed by debt. OpenAI, Anthropic, and their rivals have been issuing debt at a rate that would have drawn far more attention in other sectors and at other times.
The real source of that physical infrastructure is the issue, which turns out to be one of several. Advanced memory and processing chips are the most costly and crucial parts of AI model training. Two South Korean companies and one Taiwanese company produce the majority of them. In turn, the Persian Gulf provides the bulk of those nations’ liquefied natural gas and crude oil.
A significant portion of the energy used to power the factories that produce the chips used to train the models passes through the Strait of Hormuz. The strait is essential to almost every facet of the world economy, according to Sam Winter-Levy, a technology and national security researcher at the Carnegie Endowment for International Peace. He pointed out that the supply chain for AI is not protected. It was never the case. The business just carried on as if it were.
The fact that the dependencies were always apparent to anyone who wished to look is almost illuminating as you watch this develop. At least since the early 2020s, policy circles have been debating the geographic concentration of chip manufacturing. After the COVID-era shortages revealed the lack of redundancy in the system, Taiwan Semiconductor Manufacturing Company’s unique position in the global chip supply chain became a hot topic.
It was no secret that Taiwan and South Korea relied on imports of energy from the Gulf. The extent to which the AI investment boom was increasing risk on top of all that preexisting fragility, raising the stakes for each of those dependencies at precisely the time when geopolitical stability in the relevant regions was declining, may have been underestimated, or more charitably, purposefully not examined too closely.
There hasn’t yet been a crash due to the conflict in Iran. This distinction is important, and it’s important to avoid confusing a crisis in progress with a crisis that has been resolved. However, it is now evident that the mechanisms that analysts have been cautioning about are in action. The Gulf region’s energy infrastructure is being damaged at a rate that officials say will take years to fix. Traffic across the Strait of Hormuz has been severely disrupted. China, Japan, Taiwan, South Korea, and other Asian manufacturing economies rely on energy reserves with a limited shelf life measured in months.
The ripple effects on semiconductor production will reach Silicon Valley’s data center construction timelines with a directness that no financial engineering can completely mitigate if those reserves run out before the strait reopens to regular traffic. If build-out slows down or stops, tech companies that issued debt under the assumption of continuous exponential growth will have to make very different calculations.
To put it generously, the financial architecture that underpins all of this is tightly coupled. Formerly a senior adviser at the Consumer Financial Protection Bureau and currently employed at the Roosevelt Institute, Brad Lipton has identified structural parallels with the pre-2008 dynamics: banks lending to private credit funds, which lend further down the chain, amplifying rather than distributing the underlying risk at each stage. The AI sector raised money from investors who anticipated profits from businesses that are typically not profitable in any traditional sense.
As models become more efficient, the price of each AI-generated token—the fundamental unit of what these businesses sell—has been rapidly declining. Analysts have referred to this trend as a death spiral toward zero for the revenue models that underpin the entire investment thesis. If cost reductions are significant enough, they might eventually open up new markets and generate new sources of income to cover the expenditures. It’s also possible that the math is flawed or has never worked properly.
There is a version of this story that concludes without a catastrophic crash, in which the Strait swiftly reopens, the supply of chips stabilizes, AI companies discover sustainable revenue models, and the financial risk is distributed throughout a system big enough to spread the suffering without collapsing. A lot of things need to happen in a certain order for that scenario to work.
A longer slowdown in AI infrastructure investment as energy costs and supply uncertainty bite, a repricing of the debt that financed the build-out, and a period of reflection on whether the bet that the entire economy was covertly placed on was as sound as the pitch decks suggested make up the more likely near-term trajectory. That’s all uncertain. However, as Warren Buffett once noted, the tide can reveal who has been swimming unprotected, and at this moment, the water is running out.
