🚀 La guerre en Iran vient-elle de faire éclater la bulle IA ?

Nvidia, Alphabet, Apple, Microsoft, and Amazon collectively spent nearly $250 billion in 2025 on artificial intelligence, with $650 billion already committed for this year. Other companies like Meta, TSMC, Broadcom, Oracle, and Palantir are also investing hundreds of billions. This investment surge follows the emergence of ChatGPT in 2022, making AI a tangible reality. However, despite the excitement, AI is not yet profitable. Financial statements are often inflated by accounting tricks, and revenues come from subscriptions sold at a loss. Accounts are consolidated to mask deficits, and investments are funded by debt, not profits. The success of AI relies on three critical factors: cheap energy, fluid supply chains, and patient investors. The current crisis in Hormuz threatens this delicate balance, potentially causing the AI bubble to burst. The physical reality of AI involves complex chemistry to produce chips, requiring sterile cleanrooms, expensive machinery, and rare raw materials like helium, neon, and bromine. A third of global helium production comes from Qatar's Ras Laffan complex, which was hit by Iranian missiles in March, causing production delays of 3-5 years and doubling prices. Neon largely came from Ukraine before the war, and bromine from Israel and Jordan, creating a perfect storm of supply chain disruptions in conflict zones. Chip manufacturing is also energy-intensive. TSMC, which produces Nvidia chips for most AI applications, consumes 8-10% of Taiwan's electricity. Taiwan generates half its electricity by burning imported liquefied natural gas (LNG), a third of which came from Qatar via Hormuz. Taiwan currently has only 11 days of LNG reserves, forcing it to find alternative, more expensive suppliers. There is no easy alternative for helium, which requires specialized cryogenic transport. Furthermore, the world faces a deficit of 330,000 tons of copper this year, as data center construction for AI consumes materials intended for national grids. Transformers are being sold to the highest bidder, creating shortages. The physical capacity wall for AI has been hit sooner than expected due to the Hormuz crisis. AI companies like OpenAI are reporting massive losses. OpenAI anticipates $20 billion in revenue by 2025 but expects $14 billion in losses, not projecting profit until 2027. This requires raising an additional $207 billion. Sam Altman's claims of profitability are based on only considering inference costs, ignoring servers, electricity, salaries, data centers, and model training. Anthropic, another AI company, expects $30 billion in revenue this year but is spending $17 billion, with cash outflow exceeding inflow. They have already implemented usage quotas, limiting developers who rely on their tools. Michael Burry, known for predicting the 2008 financial crisis, identified an accounting gimmick: tech giants like Microsoft, Meta, and Amazon amortize Nvidia GPUs over four years, even though new chips render old ones obsolete in 12-18 months. These companies are extending the accounting life of their servers to 5-6 years, artificially lowering annual charges and inflating reported profits by an estimated $176 billion between 2026 and 2028. Burry shorted Palantir and Nvidia, though his timing was early. The reliance on debt, hidden losses, and inflated profits worked as long as cheap energy, fluid supply chains, and patient investors held up. However, the blockage in Hormuz has disrupted helium, copper, aluminum, and LNG supplies. Energy costs are soaring in Taiwan and the US. Investor patience is wearing thin, leading to closer scrutiny of financials. The "Magnificent 7" tech giants constitute 35% of the S&P 500, a concentration level last seen during the dot-com bubble in 2000. Half of the $6 trillion in AI spending by 2030 is still expected to be debt-financed, not profit-financed. If the Middle East situation doesn't resolve, a 30-50% correction in tech stocks could erase $33 trillion in market capitalization. A 25-30% correction in the AI sector alone would mean a loss of $4.5-5.5 trillion, a plausible scenario if Hormuz remains closed this summer. Microsoft already saw a $400 billion market cap drop in January 2026 due to impatient investors and renewed bubble fears. Meta, Alphabet, and Oracle need to raise $86 billion in debt this year, but are facing a skeptical market. For investors, rather than selling in anticipation of a crisis, staying invested and diversified for the long term is advised, as all market crises have historically rebounded. For daily AI users, it's crucial to anticipate the end of debt-subsidized AI. Monthly subscriptions costing hundreds of dollars will likely increase significantly or be limited once debt funding dries up. Exploring open-source models that run locally is recommended for those whose work depends on these tools. The debate on Wall Street continues: some, like Burry, see parallels to the dot-com bubble