121 billion dollars in bond issuances by AI hyperscalers in 2025. This sum represents approximately 330% more than the previous annual average, marking a historic shift in artificial intelligence financing. For the first time, hyperscalers are abandoning their liquidity reserves to take on massive debt, transforming AI from a self-financed technological gamble into a race where the fear of missing out supersedes economic rationality.
Goldman Sachs identifies a revealing paradox: while tech/AI stocks have outperformed the market over the recent period, their investments in this technology are exploding. This dynamic reveals that institutional FOMO is redefining the rules of technological capitalism.
The essentials
- 121 billion dollars in bond issuances by AI hyperscalers in 2025, approximately 330% more than the previous annual average
- Microsoft, Google, Amazon, and Meta are shifting from equity financing to debt for their AI data centers
- Tech/AI stocks have outperformed the S&P 500 over this period
- Goldman Sachs documents the first “FOMO debt bubble” in technological history
The hyperscaler changes its economic model
Microsoft inaugurates this mutation by issuing 22 billion dollars in bonds in January 2025, its largest debt raising since its creation. Amazon follows with 31 billion in March, Google with 28 billion in May, Meta with 18 billion in June. This synchronization reveals a structural change: AI now requires investments that even the most profitable companies cannot self-finance.
Alphabet holds 120 billion in liquidity, Microsoft 108 billion, Amazon 73 billion. Yet all prefer to take on debt rather than exhaust their reserves. This strategy reflects a conviction: AI requires such massive financial commitment that it is better to preserve strategic flexibility by resorting to borrowing, even at unfavorable rates.
AI data centers consume this windfall. A data center equipped with H100 chips costs 4.2 billion dollars compared to 800 million for a traditional center. Electricity consumption reaches 150 megawatts compared to 30 megawatts, multiplying infrastructure costs by five. These colossal investments exceed the self-financing capacity of even technology giants.
Fear of missing out on technological progress supersedes financial rationality
Goldman Sachs documents an unprecedented phenomenon: companies are multiplying their investments in a technology while their AI spending increases by 340%, illustrating how institutional FOMO is redefining technological capitalism.
This dynamic reveals that institutional FOMO is redefining technological capitalism. Corporate leaders fear less the immediate destruction of value than permanent exclusion from a future market. Jensen Huang, CEO of Nvidia, sums up this logic: “The cost of inaction now exceeds the cost of wrong investment.”
Wall Street validates this paradoxical approach. Analysts penalize companies that reduce their AI investments (-23% relative performance) more than those that go into debt to multiply them (-12%). This asymmetry confirms that financial markets are integrating FOMO as a legitimate valuation factor, upending traditional evaluation models.
The global industry facing the hemorrhaging of expert knowledge illustrates why this race is accelerating: AI talent is becoming scarce while demand explodes, forcing companies to outbid each other to avoid obsolescence.
Debt reshapes technological competition
Massive borrowing reshuffles the competitive deck. Apple, which refuses to take on debt for AI and prioritizes its 162 billion in liquidity, is losing ground to its rivals. Its AI investments reach 11 billion in 2025 compared to 43 billion for Microsoft and 37 billion for Google. This prudent restraint is costing it its dominance in intelligent services.
Tesla illustrates the opposite: the company is borrowing 15 billion specifically for its autonomous driving chips, doubling its debt but accelerating its transition to robotaxis. This strategy allows it to maintain its technological edge despite having less cash reserves than Apple.
AI debt creates a new class of technology companies. The “debt-powered innovators” accept mortgaging their balance sheets to maintain their innovation pace, while the “cash conservatives” preserve their financial soundness at the risk of technological obsolescence. This dichotomy is reshaping the competitive ecosystem.
European companies are suffering from this transformation. SAP is borrowing 8 billion to catch up on AI, ASML is mobilizing 12 billion to develop its engraving machines adapted to neuromorphic chips. Europe is discovering that it must go into debt to stay in the race against American and Chinese technology.
Economic fundamentals question sustainability
Goldman Sachs identifies three warning signals in this debt dynamic. First, the debt/EBITDA ratio of hyperscalers goes from 0.3 in 2023 to 1.8 in 2025, approaching levels historically associated with cyclical sectors. Second, their capital expenditures now exceed their operating cash flows, a situation unseen since the Internet bubble of 2000.
Third, and more troubling: the monetization of AI investments remains limited. Microsoft generates approximately 37 billion dollars in annual AI run-rate revenue. Google generated 1.2 billion dollars in Gemini revenue in 2025. This asymmetry between investment and return questions the economic viability of this race.
Goldman’s analysis reveals that 70% of current AI investments target markets that do not yet exist or remain embryonic. Companies are betting on massive adoption that may never materialize at the hoped-for level. This speculation on the future explains why debt is replacing profitability as an investment criterion.
The paradox intensifies: the more companies invest in AI, the more their operating margins erode. Microsoft sees its margin fall from 42% to 38%, Google from 29% to 24%, Amazon Web Services from 35% to 28%. AI costs more than it returns, yet companies continue investing for fear of missing the transformation.
The debt bubble redefines technological cycles
This unprecedented dynamic creates what Goldman Sachs calls the first “FOMO debt bubble.” Unlike classic speculative bubbles fed by investor optimism, this one is born from corporate leaders’ fear of being excluded from a technological revolution. Debt becomes a signal of strategic credibility rather than a risk factor.
Traditional technological cycles followed linear logic: research, prototype, commercialization, adoption, maturity. AI inverts this sequence: companies go into massive debt before even mastering the technology or identifying its commercial outlets. This inversion transforms innovation into a financial race where borrowing capacity determines competitive position.
Wall Street anticipates a tipping point in 2026-2027, when the first AI debts come due without having generated the expected revenues. Goldman Sachs models three scenarios: soft landing if AI generates 15% of promised revenues, correction if it reaches 8%, crisis if it remains below 5%. Currently, it hovers around 3%.
This transformation could permanently reshape the technology economy. Companies that survive this phase of massive debt will have dominant AI infrastructure for the following decade. Those that collapse under their debt burden will free up a market that survivors will monopolize.
AI is entering its riskiest phase: one where the scope of financial bets exceeds understanding of technological stakes. The debt of hyperscalers reveals that technological innovation now obeys less rationality than the psychology of fear of missing the future.