The AI Bubble: Has It Burst, Or Not Yet?
AI in the age of polycrisis. An illustration through chips manufactured in Taiwan.
For nearly a year now, many voices have been diagnosing a speculative bubble¹ around the generative AI industry. Sam Altman, head of OpenAI, conceded as much himself in August 2025. The sector is literally seething, caught between stratospheric stock market valuations and colossal investments — and therefore, at times, colossal debt.
Taking on debt for several years before reaching profitability is nothing new in tech. Amazon lost $3 billion between 1995 and 2003 before turning a profit²; Uber — the most recent reigning champion — lost $33 billion over 13 years³.
OpenAI expects to lose $115 billion between 2024 and 2028 and to become profitable in… 2030. Mr. Altman has stated that he would need the trifling sum of $7 trillion to develop the future of AI. That’s twice France’s GDP, or 20 years of the world’s budget for dealing with current and future climate crises and disasters⁴.
The profitability equation for generative AI is nevertheless extremely difficult to solve. Massive and very costly data centers are springing up at breakneck speed to handle the influx of users and the ever-hungrier AI models. The chips that perform the calculations have to be replaced every 2 to 3 years. And all of this without economies of scale: each new user requires more computing power.
Numerous circular investments are being made in the sector — Nvidia investing in OpenAI, for example, while OpenAI in turn buys chips from Nvidia. It only takes one player to cough for the entire ecosystem to catch the virus, which can quickly trigger cascading effects.
What’s more, the cluster can quickly turn into a pandemic. AI companies are taking on massive debt from asset managers⁵ like BlackRock, which in turn have raised funds from major banks. An AI bubble that bursts could set off a chain reaction and spread into a major economic crisis.
While reaching profitability already amounts to a high-wire balancing act in “normal” times, many factors could make the exercise even more perilous. Between the energy, climate, and geopolitical crises — welcome to the age of polycrisis, which makes the AI bubble even more fragile. Let’s illustrate this with a focus on chips manufactured in Taiwan.
Chips: The Sinews of War
AI models, however virtual they may seem, need very real infrastructure to function. To train and run AI models, ever-increasing volumes of calculations must be performed by chips.
For an Nvidia H100 or H200 GPU chip designed for generative AI, expect to pay between $25,000 and $40,000 apiece. Tens of thousands of them can be found in the colossal data centers erected since the generative AI boom. They make up the bulk of the cost of training the models.
Many risks weigh today on the cost of these chips. Energy, climate, and geopolitical crises can go as far as causing stock shortages. The sector’s heavy concentration exacerbates these threats: 90% of the chips needed for AI are manufactured by the company TSMC⁶, based in Taiwan.
OpenAI made its profitability forecasts on the basis of relatively stable energy costs — and therefore chip costs. But if those costs were to explode, how long would investors keep their faith in the sector?
Energy Crisis
“The war in the Middle East is generating the largest oil supply disruption in history.” — Fatih Birol, Director of the International Energy Agency (IEA), March 16, 2026
Oil is the lifeblood of our economy today. From the diesel in trucks to the lubricants in industrial machinery, by way of the plastics derived from petroleum, any good or service in our economy depends on it to be produced. Without the continuous resupply of Paris by truck, stores would have no food left to sell within five days. Less than a week.
The blockage of the Strait of Hormuz has halted the supply of 20% of the world’s oil. To return to the bloodstream image, this is like pressing on our economy’s carotid artery⁷. Several international organizations, including the International Monetary Fund (IMF) and the World Bank (WB), warned on Friday, May 29, 2026 of a risk of oil shortages this summer if maritime traffic through the Strait of Hormuz does not quickly return to normal.
Even once the blockage ends, several years will be needed for the oil and gas industry to return to a normal supply rhythm. The IEA’s director, Fatih Birol, speaks of the worst energy crisis in history. Worse than the oil shocks of the 1970s, worse than 2008, worse than 2020.
Once again, every industrial process is dependent on oil. To put it simply: when the price of oil climbs, all costs climb. This situation suggests, at best, price increases across the entire AI value chain — from chips to data centers — and, at worst, an unprecedented financial crisis.
Moreover, most of the oil from the Strait of Hormuz is exported to Asia, including Taiwan and South Korea. These two countries supply the vast majority of the chips needed for AI data centers. Beyond oil and gas, AI chip manufacturers source helium, sulfur, and bromine — resources that come mostly from… the Strait of Hormuz. Since these are already the most expensive components in training AI models, chip prices had better watch out.
Climate Crisis
“These days, whenever I pass by a temple or a palace, I never fail to offer a sincere prayer for favorable weather.” — Tsai Ing-wen, President of Taiwan, during the 2021 drought, the worst in 56 years
Taiwan is not spared by climate disruption. In 2021, the country suffered its worst drought in 56 years. TSMC had to cut its water consumption by 10% and resupply its main plant with tanker trucks. This forced slowdown in activity worsened the global shortage of electronic chips.
Yet chip foundries need more and more water as demand explodes. Future droughts could weaken the AI industry, ranging from a “mere” increase in chip prices to an outright shortage of them.
Geopolitical Crisis
“The question of Taiwan is the most important in Sino-American relations. If it is handled well, relations between the two countries can remain broadly stable. If it is handled badly, the two countries will clash, or even enter into conflict.” — Xi Jinping, President of China, May 14, 2026 (the word “conflict” used in Mandarin does not necessarily mean a military conflict)
China has never hidden its ambition to take control of Taiwan. Beijing says it favors a peaceful option, without, however, ruling out armed intervention. For several years now, the People’s Liberation Army has been organizing increasingly large military exercises around the island of Taiwan. These shows of force raise fears of a risk of military escalation, which would have disastrous consequences for the availability of AI chips.
The Taiwanese semiconductor industry is, however, regarded as a “silicon shield” for the island, with the Chinese and Taiwanese economies being particularly dependent on each other. 50% of Taiwan’s electronic products are exported to China⁸.
In order to get around this silicon shield, China is pushing to gain technological sovereignty over semiconductors, particularly in the AI industry. Achieving such a goal could still take years. But if Beijing pulls off its gamble, invading Taiwan would affect it far less than it would today.
Final Word
Manufacturing chips for AI demands a very high level of expertise that takes years to attain. TSMC will remain, by far, the leading supplier to the AI industry for a long time to come. The energy, climate, and geopolitical risks inherent to the Taiwanese context are therefore to be taken seriously.
And yet, these factors are only a sample of all the threats weighing on the AI bubble. In early March, Iran bombed Amazon data centers in the United Arab Emirates. Donald Trump’s actions are often unpredictable — the latest having triggered the worst energy crisis in history. The financial health of the asset managers that have invested heavily in the AI industry is fragile. It might be hard for the new tech moguls to get a good night’s sleep sometimes.
Benjamin Karas, retraining at 42 Paris, co-editor of The Turing Point newsletter.
Note on the use of AI: some background research was conducted with the help of the AI Perplexity. The structuring of ideas and the writing in French were done without AI assistance. The translation to English was realized with the help of Claude AI.
This article may not be reproduced or used for the purpose of training artificial intelligence systems. Text and data mining is prohibited in accordance with Article 4(3) of Directive (EU) 2019/790.
To go further
For French speaking readers, feel free to read “AI — The Great Smokescreen” (“IA — Le Grand Enfumage”), co-written by Lou Welgryn and Théo Alves Da Costa. Many thanks to them; their book was invaluable to me during the background research for this article.
Notes and sources
¹ A significant rise in the prices of financial assets (stocks and bonds) or in prices within a given sector (commodities, real estate, etc.) without the actual economic situation justifying it. The phenomenon accelerates: the more people buy, the more their neighbors buy so as not to miss out on the rise, and so on. Demand is artificially sustained. Then, at a certain point, some buyers get spooked: they become sellers and the spiral reverses. Prices fall more or less quickly. The bubble is said to “burst,” or this is also referred to as a “crash.” (French Ministry of the Economy)
² Amazon.com, “Annual Report on Form 10-K for the fiscal year ended December 31, 2003,” SEC, February 25, 2004
³ Uber Technologies, “Annual Report on Form 10-K for the fiscal year ended December 31, 2022,” SEC, February 21, 2023
⁴ UNEP, “Adaptation Gap Report 2023,” UN Environment Programme, November 2, 2023
⁵ Asset management consists of managing investment portfolios on behalf of a variety of clients: individuals, companies, or institutions such as banks and insurance companies. The asset manager’s mission is to grow the sums invested, taking into account each client’s specific objectives, their financial needs, and their risk tolerance. (French Asset Management Association)
⁶ TSMC — Taiwan Semiconductor Manufacturing Company
⁷ Thanks to Matthieu Auzanneau for this image. Mr. Auzanneau is an author specializing in ecology and economics, particularly energy issues. His recent appearance on the Sismique podcast analyzes in detail the consequences of a blockade of the Strait of Hormuz.
⁸ Taiwan Bureau of Foreign Trade; Office of the United States Trade Representative
