The immense economic and ecological risks being taken by the artificial intelligence industry have grown so impossibly large that no one — including the AI companies — has the means to gauge them. This historic boom, like so much else in AI , is run purely on vibes. In every direction, AI companies are straining to expand beyond their capacities in three key areas: industrial supply chains, grid electricity capacity and global capital markets. High-tech companies occupy a world of structures, protocols and mutual interests that requires guaranteed supplies of rarefied parts and materials to be delivered with precision. If energy and mineral supplies cannot be guaranteed, if capital is no longer liquid and if long-term commitments cannot be met, then that world rapidly unravels. The tech billionaires talk excitedly about “ existential risk ,” but it is abundantly clear that none of them has any conception of systemic risk — the profound dangers that arise when vast complex systems impact one another in unforeseen and uncontrollable ways. But this ignorance cannot continue much longer. Even as AI CEOs continue projecting otherworldly confidence in near-term “10x” growth, the cracks in their world-bending visions are beginning to show. The term “bubble” does not do justice to the gravity of the situation; a failure of AI will be less like a burst than a systemic collapse. The term “bubble” does not do justice to the gravity of the situation. The U.S.-Israeli attack on Iran has only accelerated these entropic dynamics. The world now faces the possibility of an energy crisis, a supply chain disruption and a massive economic contraction, each with profound impacts on an AI industrial complex dependent on fossil fuels from the Persian Gulf. “The vase is broken, the damage is done,” said Fatih Birol, head of the International Energy Agency, from his organization’s Paris headquarters in late April. “It will be very difficult to put the pieces back together.”
The immediate impact of the Iran war is not simply in fossil energy, but also on a huge range of fossil fuel byproducts on which AI and many other industries depend. The global economy is still very much a fossil fuel economy, and AI is locking us even further into it.
Consider helium. A byproduct of natural gas, the element is irreplaceable for cleaning and treating silicon wafers. “Without it, you can’t make the product whatsoever,” said Aqib Zakaria, an analyst at ChinaTalk, a think tank. “It has a disproportionate importance relative to its economic value.”
Thirty percent of the world’s helium supply was produced at Qatar’s Ras Laffan Refinery. That has now stopped, and after an Iranian drone attack it will take years to fully repair. While stockpiles of helium were high, the supply chains are complex and run with no contingency. Helium is only one threatened input. The entire AI supply chain is a hyper-fragile system of critical choke points now struggling with “an ongoing fragmentation of global supply chains,” said S. Yash Kalash, a senior fellow at the Centre for International Governance Innovation.
Further up the supply chain, Indonesia, one of the world’s largest mineral exporters, is cutting back mining of copper, nickel and silver, all essential for AI microchips and infrastructure. (AI data centers require an estimated 30,000 tons of copper for every gigawatt of power.) The cutbacks are due to shortages of sulfur, which is used in many mining processes, as well as oil and gas. “When one component is no longer available, all of a sudden the whole supply chain goes down, and you have a type of collapse,” said Craig Tindale, a former infrastructure executive at Oracle who closely monitors the tech supply chain. These systemic dependencies reveal a key point: While AI superficially appears like another digital platform industry — providing services and products like ChatGPT and Claude — it is really more like a heavy industry. Every new AI user requires more computing power — what AI companies call “compute” — that in turn requires more energy and resources. Between 2010 and 2022, despite the extraordinary growth of big tech platforms, data center energy demand remained virtually static as those companies switched to more efficient “hyperscale” data centers. That all changed following the launch of ChatGPT; since 2022, energy demand of data centers has more than doubled. The International Energy Agency expects energy demand to more than double again by 2030 in its “base case” forecast. With AI, there are no economies of scale. The entire AI supply chain is a hyper-fragile system of critical choke points. The paradox of a heavy industry attempting to grow at the speed of a social media startup has created an intractable tension that is pulling the hastily assembled AI industry apart at the seams.
Recently, Google’s head of AI infrastructure, Amin Vahdat, was asked which shortage was causing the most serious bottleneck of the company’s data center expansion plans. Was it a lack of labor, AI graphics processing units or power?
“All three of those are major, major issues for us,” responded Vahdat. “I’m not relaxed about any of them. I couldn’t pick one.”
Perhaps most alarming is the industry’s staggering power needs. The energy shock of the Iran war has yet to hit the U.S. data center build-out, but the industry’s incredible energy demands are already breaking U.S. power grids.
Energy analyst Grid Strategies recently estimated that data centers will require 90 gigawatts of new power in the U.S. alone by 2030, or about 25% of all U.S. electricity demand today. But only 24 gigawatts of new electricity capacity is being added to the U.S. grid over that time. In effect, AI companies are building their own power, nearly all of it using planet-warming natural gas. There are now long bottlenecks on gas turbines and electrical transformers. As a result, many data centers are now using converted jet engines, which are twice as polluting as purpose-built turbines.
Despite being gas powered, data centers also require massive, grid-scale batteries to even their power load, causing delays in their rollout to other parts of the grid. In this way, AI has slammed recent progress made on decarbonizing electricity grids into reverse.
In 2024, OpenAI CEO Sam Altman mused about a not-too-distant future where “intelligence was too cheap to meter.” This vision required building AI models at an unfathomable scale, with no regard to the cost, energy or resources used. It was assumed that if usage rapidly grew, like other digital platforms, it would all pay off eventually.
But the costs of using AI continue to soar, despite AI companies spending $720 billion in 2026 alone. As a result, the AI data center build-out has slowed to a crawl. The research group Data Center Watch reports 48 data center projects were blocked in 2025. Another research group, Sightline Climate, estimates that of 16 gigawatts of data centers planned for 2026, only 5 gigawatts are actually under construction. Much of this gap is likely due to power bottlenecks.
As a result of the slowdown, AI companies don’t have enough capacity to serve their AI models that depend on prodigious amounts of compute and energy. And this is now an economic problem for the industry that is edging toward existential. While AI companies never cease to boast about futures of exponential growth, underneath the bravado is a terrifying problem: Their losses are going up faster than revenues. Much faster. Open AI is hemorrhaging more money than any company in history. Early last year it forecast a “burn rate” of $35 billion until 2030. Since then, it’s adjusted that forecast twice, and now the burn rate is up to $228 billion. OpenAI will lose $85 billion in 2029 alone — the biggest ever corporate loss in a year — before (it predicts) miraculously turning a profit of $45 billion in 2030. The costs of using AI continue to soar, despite AI companies spending $720 billion in 2026 alone. Despite raising billions in funding rounds, the financial doom loop appears intractable. OpenAI recently missed its revenue targets, while its chief financial officer was reported as saying that it might “not be able to pay for future computing contracts if revenue doesn’t grow fast enough.”
All AI companies are facing the same problem, but have so far not adjusted to this reality. Indeed, it’s hard to see how they can. Having sucked in hundreds of billions of dollars from financial markets on the premise of scale, they can hardly pivot now.
This represents another departure from previous digital platform eras, when big tech companies used to accumulate vast mountains of cash. In AI’s financial bubble, the sector has become by far the largest issuer of debt on global finance markets. The amounts are in the trillions. The Financial Times recently forecast that $9 trillion will be spent on AI infrastructure by 2030. Adjusted for inflation, this would pay for over 270 Manhattan Projects. Over $1.2 trillion in corporate bonds has been issued to AI-related companies in the past two years alone. U.S. banks, meanwhile, are anxiously trying to offload their exposure to the massive debt. As one banker confided to the Financial Times, “The sizes we’re talking about [are] out of scale to anything we’ve thought about, ever.” The swirling mass of interconnected multibillion-dollar deals, often called AI’s “circular economy,” links all AI companies together in one inextricably entangled speculative frenzy. The upshot is that AI companies, despite their courtroom battles, their feuds and their hand-holding tiffs, are all entirely codependent; if one goes down, they all go down.
And if they go down, so much of global finance is now so deeply invested in these companies that the shock waves of the failure will be seismic.
The conflict in the Gulf has compounded the AI industry’s problems in more ways than one. The enormously wealthy sovereign wealth funds of the United Arab Emirates and Saudi Arabia have been huge investors in OpenAI, Anthropic and Elon Musk ’s xAI, and plans were in the works for huge AI data centers in Saudi Arabia and the UAE, where (fossil) energy is abundant and public resistance is nonexistent.
But these countries — now with broken supply chains, severely damaged infrastructure and their image as tax-free havens of stability now in ruins — are facing their own existential challenges. Iranian drone attacks on three Amazon data centers in the region have put all AI projects on extended hold and may well hold back further investment. Research group Epoch AI recently rated Gulf states’ withholding of capital investment a “high risk” for the AI boom. “What this war has shown is that the United States can’t guarantee any security in the region, given how many attacks have landed on these countries,” said Kalash, the fellow with the Centre for International Governance Innovation. If one of these company’s IPO fails, it will have a cascading impact through the tech sector. The loss of Gulf states’ capital risk IPO plans by SpaceX, now merged with xAI, OpenAI and Anthropic. All these companies urgently need to go public, as it’s the only place they have left to raise the vast amounts of capital they need. SpaceX plans to go first in June. Its entire pitch to investors hinges on launching its data centers in space for xAI. OpenAI plans to follow soon after, but its chief financial officer, Sarah Friar, has warned that “company isn’t yet ready to meet the rigorous reporting standards required of a public company.” Hardly reassuring for a company worth $875 billion.
Any one of these IPOs — each expected to be worth between $50 billion and $100 billion — would be by far the largest in history. Many analysts question if the financial markets can swallow three AI elephants in one year. And if one of these company’s IPO fails, it will have a cascading impact through the tech sector. Considering the wider economy’s exposure to tech — whose seven biggest firms account for 30% of the stock market — this is unlikely to be shrugged off like the dot-com bubble.
The Bank of England recently warned of the systemic risks deeply embedded in the AI bubble. “Valuations for U.S. technology companies focused on AI remained particularly stretched,” it stated in the Financial Policy Committee meeting record from March. “AI-related repricing could transmit widely throughout the financial system and impact the real economy.”
The modern world has instilled in us all the expectation of inexorable progress. In 2026, AI is the ultimate — and perhaps only — remaining symbol of this idea. Despite the widespread fear that AI provokes, the fear of stopping it, of stopping progress, is for many of us an even more frightening prospect. Maybe this is what drives the AI boom ever onward, however much chaos it creates.
But as AI drags us toward energy, economic and geopolitical tumult, Silicon Valley’s platitudes of AI-enabled progress seem all the more detached from reality. The narrative of progress appears to be crumbling like the supply chain for a high bandwidth memory chip. When our narrative of progress stutters and fails, a narrative of collapse swiftly fills the void left behind.
In this narrative, AI can be seen as a pan-global collapse machine. All likely scenarios of AI’s future contain at least some dimension of collapse. The AI bubble will either collapse in on itself, causing systemic chaos across the global economy, or AI will reach some kind of terrifying inflection point, where it presents an array of ecological and social “existential threats.” But even its “success” could take the form of a collapse: Ninety-nine percent of the human economy becomes a permanent underclass, eking out survival in a ravaged world, while AI and its overlords spiral merrily onward.
Whether you are for AI or against it, it’s collapsing all the way down.
The post $9 Trillion Collapse Machine appeared first on Truthdig .