Deutsche Bank: Market Underestimates the Risk of an AI Crash and a Flight to Bonds

Deutsche Bank strategist George Saravelos believes investors have grown too disillusioned with bonds in 2026. In his view, the market underestimates the risk of an "AI blowup": a sharp repricing of AI assets that would send capital running into bonds. Bloomberg reported this on October 9. The published summary is short, so below we walk through the logic of the thesis rather than figures from the bank's full note.

What Saravelos Is Actually Arguing

The first part of the thesis concerns sentiment. Over the past year, positioning in fixed income has swung deep into negative territory, and Deutsche Bank considers that an overshoot. When nearly everyone agrees that bonds are not worth buying, prices have usually already absorbed the bad outcome.

The second part concerns a risk that is not in those prices: an abrupt break in the AI trend. That would mean falling market capitalizations, revised capital-expenditure plans and doubts about whether the infrastructure will pay off. In that case, investors would start leaving risky assets, and bonds would become one of the main destinations for that flow.

Key elements of the argument:

• the bank judges that sentiment on bonds in 2026 is overly negative;
• the market underestimates the probability of a sharp correction in AI;
• if such a correction happens, demand for bonds as a safe haven will rise;
• the asymmetry favors bond buyers: the negative scenario is not reflected in prices.

Why AI Ended Up at the Center of the Bond Story

A couple of years ago, a link between neural networks and the government debt market would have looked like a stretch. Now it is quite natural. A significant share of the U.S. stock market's growth rests on a handful of large technology companies that are spending enormous sums on chips, data centers and power supply. Part of that spending is financed with borrowed money: in 2025, hyperscalers and related companies issued bonds worth tens of billions of dollars.

AI therefore affects the bond market from two sides. As a borrower, it increases the supply of corporate debt. As a driver of equities, it shapes investors' appetite for risk. If the AI rally stops abruptly, the stock market loses its main locomotive, and the usual reaction, a move into safe assets, could turn out stronger than market participants currently assume.

Where the Argument Is Vulnerable

The thesis has weak spots. "Bonds" are not a single asset. In an AI-driven crisis, government bonds could rally while the corporate debt of the AI borrowers themselves falls as spreads widen. A buyer would have to distinguish between these segments.

The second caveat: bearish sentiment on bonds has its own causes, including budget deficits, inflation risks and uncertainty over policy rates. An AI failure would not cancel them out. It could add demand for protection, but it would not necessarily override the other factors.

Finally, Saravelos himself is talking about a risk, not a forecast. This is hedging logic: an assessment that the insurance is cheaper than it should be.

How to Read This in Practice

For investors, the takeaway is this: a portfolio built entirely on AI growth gets natural cover from high-quality bonds, and Deutsche Bank thinks that cover is currently cheap. But it is worth looking at maturity, credit quality and correlation with the tech sector.

For technology companies and the finance chiefs of AI projects, the signal is different. The capital market is looking more closely at how much AI infrastructure depends on debt and what revenue it is counting on. The more transparent the payback model for data centers and compute contracts, the lower the chance of ending up at the center of a repricing.

One last point. When strategists at major banks start describing an AI failure as a macro scenario rather than a single-sector story, it shows how deeply the theme is embedded in the broader market. Watching it now means following both Nvidia's earnings reports and the yield curve.