Speaking at the Global Financial Leaders' Investment Summit in Hong Kong on Tuesday, the man who runs PIMCO said the sentence the industry has been half-listening for. There will be a downturn in the AI market, CEO Emmanuel Roman told the room, "and then things are going to break."
The warning matters less than its source. PIMCO, the legendary bond house, has itself been one of the principal financiers of the AI boom: it was among the group behind the first chip-based financing for CoreWeave in 2023, it made a quick $2 billion in paper gains on $29 billion of Meta bonds meant to support the buildout, and it bought $4 billion of SpaceX bonds last month. When the head of that firm publicly sounds the alarm — having just sold a chunk of its AI debt — the market tends to notice.
And the market keeps going. Meta drew $58 billion in orders for a bond offering expected to raise $15 billion to $20 billion, Bloomberg reported. Oracle said on Monday it would need another $20 billion in bonds on top of its existing commitments. Alphabet is selling €6 billion (about $6.7 billion) of bonds. Even Microsoft came to market recently, for the first time in a decade, with $25 billion.
Roman's arithmetic is the part worth sitting with: the AI buildout could create $2 trillion in financing needs over the next five years. The entire US bond market is about $12 trillion.
Apollo Global Management, which has lately been in talks with SpaceX about a $40 billion chip deal, is also counselling caution. President Jim Zelter told analysts that society will use "dramatically more computing power in the future" and that AI is improving quickly — but that significant uncertainty remains. "What we don't know is how pricing and economics will evolve."
The sheer scale of demand for Meta's paper suggests this is not the top of the market, or not yet. Reports from China suggest its own AI bubble burst some time ago. Meanwhile Wall Street banks have been tripping over each other to finance $63 billion of debt for Anthropic — not even a hyperscaler.
What has changed is the unanimity. For two years the trade rested on a shared assumption that compute demand would outrun any constraint; now the most exposed lenders are the ones describing the exit conditions. Roman named US rate hikes, war and a eurozone crisis as plausible triggers. The year already produced one cautionary tale in the collapse of Situational Awareness.
Nothing here predicts a date. It predicts that when the turn comes, it will arrive through the credit channel first — and that the people now warning about it are the same people who built the market.
