Wall Street is starting to question how the AI buildout gets paid for. At the 2026 Global Financial Leaders’ Investment Summit in Hong Kong, PIMCO chief executive Emmanuel Roman warned that the AI market will take a hit. “And then things are going to break,” he told the audience.
Roman’s concern is not the technology itself. It is the financing stack underneath it. He estimates the AI buildout could need $2 trillion in financing over the next five years. Nobody knows how pricing and economics will evolve as that debt matures.
The bond market is already absorbing a remarkable volume of issuance. Meta drew $58 billion in orders for an offering expected to raise $15 billion to $20 billion. Oracle said Monday it needs another $20 billion in bonds, adding to existing commitments. Alphabet is selling 6 billion euros, or roughly $6.7 billion. Microsoft sold $25 billion in bonds, its first such sale in a decade.
Demand for these deals remains strong, which complicates the bubble argument. Investors keep buying AI-linked debt even as executives warn about risk. That gap between caution and appetite is where trouble usually starts.
Much of the money funds data centers, chips, and power contracts. Those assets take years to generate returns, while the bonds carry fixed schedules. A slowdown in AI revenue would hit companies that borrowed against future demand.
Neocloud providers face the sharpest test. They rent capacity to AI labs and fund hardware with debt. Microsoft has reportedly begun turning off rented capacity from Oracle, partly because the hardware used older Nvidia chips. A single cancelled contract can ripple through a lender’s book.
None of this means the buildout stops. Meta, Alphabet, and Microsoft still report strong cash flow and can service their obligations. The risk sits with smaller operators and with the lenders who financed them at thin margins.
Roman’s warning lands at an awkward moment for Anthropic and OpenAI. Both companies are chasing enormous valuations while their compute bills climb. Public markets will eventually ask for a clearer path to profit.
For now, the signal is simple. AI companies can still borrow, but the terms matter more than the headlines. Watch the spreads, not the speeches.
Source: The Verge — Please stop asking me when the AI bubble will pop.
Related reading: Neocloud Firms Face a Debt-Fueled Test as AI Bubble Fears Grow and TSMC Weighs Entering the RAM Business as AI Demand Squeezes Global Memory Supply.
The Netherlands, for internationals
One email a week: the week's news for internationals in the Netherlands, new practical guides and what changed in the cities. No spam, unsubscribe any time.
