Market consensus on the U.S. economy has undergone a complete reversal within a single year, and the most underestimated systemic risk may lie not in the bond market or in fiscal policy, but within the AI ecosystem itself.
According to Deutsche Bank's global head of foreign exchange research George Saravelos, who published a report after completing an intensive round of client visits along the U.S. East Coast, last year's dominant narrative was that AI would prove deflationary and that the U.S. Treasury would never allow a bond selloff. This year, the narrative has flipped entirely: AI-related debt issuance is now seen as a major force driving yields higher, and the U.S. Treasury has lost control over long-end interest rates.
He also noted that discussions have emerged about pausing 20-year Treasury issuance and significantly shortening the duration of supply, though most clients remain extremely cautious on fixed income assets.
Saravelos disagrees with this sentiment. In his view, the narrative pendulum has swung too far, and the most underpriced tail risk in the market today is a breakdown within the AI ecosystem itself 鈥?whether through a safety incident, a failed IPO, or disappointing revenue. Should this risk materialize, it would place significant pressure on the dollar while providing strong support to the bond market, yet current market pricing reflects almost none of this possibility.
Narrative Reversal: From AI Taming Inflation to AI Driving Yields Higher
According to the Deutsche Bank report, a year ago Wall Street's prevailing consensus rested on two mutually reinforcing judgments: first, that AI technology would suppress inflation by boosting productivity, and second, that the U.S. Treasury was capable of managing the bond market and that long-end rates would not spiral out of control.
That framework, however, has been completely abandoned by the market over the past year. The current mainstream view holds that large-scale bond issuance by AI-related companies is becoming a significant force pushing Treasury yields higher, and that the U.S. Treasury 鈥?whether through Bessent or other policy tools 鈥?is powerless to suppress the rise in long-end rates.
Saravelos recorded a specific signal circulating in client meetings: discussions about halting 20-year Treasury issuance and substantially reducing the overall duration of supply. This rumor itself reflects deep market anxiety about the supply-demand imbalance at the long end. Nevertheless, Saravelos made clear that he personally believes the narrative pendulum has swung too far, and that current pessimism toward fixed income may represent an overcorrection.
The France Problem: A New Source of Pressure on the Euro
During these client visits, the France issue formed another important thread of discussion. Saravelos noted that client sentiment on France was overwhelmingly bearish, with a tone similar to economist Paul Krugman's recent assessment, tending to draw parallels between France's current fiscal predicament and the European debt crisis period of 2010 to 2015.
Deutsche Bank takes a different view. Saravelos said he explained in meetings why he disagrees with that analogy, but acknowledged that given the significant dislocations seen in the market last week, restoring market confidence will take time. He characterized the France issue as a new source of euro pressure that was not originally expected to emerge this year, suggesting this factor will continue to weigh on the euro in the near term.
The Real AI Risk: Data Center Strain and Questionable Revenue Logic
On the AI topic, Saravelos described an unexpectedly deep discussion. He participated in a panel of data center industry experts, and participants broadly expressed concerns on three fronts: underwriting standards are declining, uncertainty exists about whether energy supply can meet construction demand, and it remains unclear whether the expected massive expansion of inference computing capacity can secure sufficient demand and revenue support.
The report cited a Brookings Institution research paper stating that AI labs would need to generate nearly $4 trillion in revenue to achieve returns on data center investments. This figure provides a quantitative reference point for the sustainability of the current AI investment boom.
The Most Underestimated Systemic Risk: A Potential Collapse of the AI Ecosystem
Saravelos explicitly identified what he considers the market's most dangerous blind spot. He wrote that the biggest systemic risk for the market next year is not France, but something going wrong within the AI ecosystem 鈥?which could take the form of a safety incident, a failed IPO, or a disappointing revenue report.
He emphasized that concentration risk in the AI space is extremely high. Should such a risk event occur, its market impact would be a significantly negative shock to the dollar while providing strong positive support to the bond market. However, Saravelos judges that this event risk is currently severely underpriced in the market.
This assessment stands in stark contrast to the prevailing market narrative 鈥?while the market broadly worries that AI debt issuance is pushing yields higher, Deutsche Bank is flagging a tail risk in the opposite direction: the unraveling of the AI narrative itself.
Comments