PIMCO's Roman Warns AI Market Faces Downturn: "Things Are Going to Break"

By Marcus Bennett

Emmanuel Roman, CEO of Pacific Investment Management Co. (PIMCO), the bond manager whose firm has been instrumental in financing the AI buildout, told an audience of financial executives in Hong Kong that a downturn in the AI market is coming — "and then things are going to break."

Roman delivered the warning at the 2026 Global Financial Leaders' Investment Summit, according to a Bloomberg roundup of the event titled "Wall Street CEOs, Big Money Managers Divided Over AI Risks." His firm has estimated that the AI infrastructure buildout could generate $2 trillion in financing needs over the next five years. For scale: the entire US bond market runs about $12 trillion.

The warning matters because of who is delivering it. PIMCO was among the financiers that completed the first CoreWeave chip-based financing deal in 2023, a structure that used AI chips as collateral and helped inaugurate the current wave of AI-linked debt. Last year the firm booked a quick $2 billion in paper gains from $29 billion in Meta bonds issued to support the AI buildout. Last month it bought $4 billion of SpaceX bonds. If the CEO of that firm is publicly predicting a break, and his company has recently sold off a portion of its AI debt, investors are reading it as an insider's signal.

How Much Debt Is Actually Piling Up?

The astronomically expensive AI buildout has swamped debt markets, making it harder for issuers to raise cash. The issuance keeps coming anyway, and buyers keep showing up.

• Meta drew $58 billion in orders for a bond offering expected to total $15 billion to $20 billion, Bloomberg reported.
• Oracle said Monday it needs another $20 billion in bonds, on top of its existing steep commitments.
• Alphabet, Google's parent company, is selling €6 billion — roughly $6.7 billion — of bonds.
• Microsoft recently sold bonds for the first time in a decade: $25 billion worth.

Roman has put a number on the total exposure. His estimate of $2 trillion in AI financing needs over five years would represent roughly one-sixth of the entire US bond market, a concentration level that gives fixed-income managers reason to pause.

The order book on Meta's offering cuts the other way. A $58 billion bid for a $15 billion to $20 billion deal indicates demand still far exceeds supply, which suggests the market has not reached its top.

What Did Apollo Say?

PIMCO is not alone in urging caution. Apollo Global Management — the firm reportedly in talks with SpaceX about a $40 billion chip deal — is also flagging AI risk.

Apollo president Jim Zelter told analysts that while he expects society will use "dramatically more computing power in the future" and that AI capabilities are improving quickly, significant uncertainty remains.

"What we don't know is how pricing and economics will evolve," Zelter said.

That sentence is the crux of the debate. Nobody disputing AI's technological trajectory. The open question is whether the revenue from that computing power will service the debt now being raised to build it.

What Are the Warning Signs?

The signals right now point in opposite directions.

On the bullish side: Meta's oversubscribed bond sale, and Wall Street banks competing to finance $63 billion in debt from Anthropic — a company that is not even a hyperscaler (the term for the largest cloud and AI infrastructure operators such as Microsoft, Amazon and Google). Banks tripping over themselves to lend $63 billion to a frontier lab indicates appetite remains fierce.

On the bearish side: reports from China indicate its AI bubble burst some time ago, offering a case study in how quickly the cycle can turn. And the internal dynamic of the market has shifted. As more participants express caution — particularly after what observers describe as the disaster at Situational Awareness — the most euphoric phase of the cycle may already have passed.

What Could Trigger a Break?

Roman singled out three potential triggers for a market flip: US rate hikes, war, or a eurozone crisis. Each would raise financing costs or choke risk appetite at exactly the moment the AI buildout needs continuous access to credit.

The arithmetic is unforgiving. A $2 trillion financing program over five years assumes debt markets stay open and cheap. A macro shock that pushes yields higher, or a geopolitical event that freezes issuance, would hit leveraged AI infrastructure projects first. Those projects carry long payback periods and depend on collateral — chips, data centers — whose resale value is untested in a downturn.

The market is not currently in what analysts would call go-go-go territory, a description of peak speculative momentum. That could change. One plausible catalyst: Anthropic's IPO. If it prices as a blockbuster, the effect could resemble someone arriving at a party that is running low on beer with a 24-pack — a fresh jolt of enthusiasm precisely when the mood was fading.

Why Nobody Can Call the Timing

The honest answer to the question of when the AI bubble pops is that nobody knows — and anyone who did know would not share the information freely. A forecaster with genuine timing knowledge could make a life-changing leveraged bet on the market, retire on the proceeds, and never write another word.

What is observable is the structure of the market: a PIMCO CEO who built his firm's AI position now publicly warning of breaks; an Apollo president questioning pricing and economics; hyperscalers and frontier labs raising debt at a pace approaching $2 trillion over five years; and order books still deep enough to absorb it all. The tension between those facts is the story.

The composition of the buyer base will determine the outcome. If demand for AI-linked debt keeps pace with the $2 trillion issuance pipeline, the buildout continues on schedule. If a rate hike, war or eurozone crisis drains that demand, Roman's forecast — a downturn, then breaks — moves from warning to playbook.