AI Model Release Does Not Significantly Lower Interest Rates or Alleviate Fiscal Concerns

Overview

This UBS Global Strategy report explores the potential impacts of artificial intelligence (AI) development on interest rates and fiscal risks. The report uses an event study method to examine how the yields on US government bonds and the market's assessment of fiscal risks react when leading AI labs release advanced models. The findings indicate that from 2025-26 onwards, the release of AI models did not significantly lower interest rates, and the market did not substantially adjust its pricing of US fiscal risks because of these releases. Overall, while AI holds promise for long-term productivity gains, it is more likely to exert inflationary pressure in the short term, making fiscal and monetary policy combinations the primary drivers of interest rates.

Core views

The effect of AI on interest rates is a topic of ongoing discussion among academics and investors. There are two opposing theoretical frameworks: one suggests that AI-driven productivity gains enhance consumers' optimism about future growth, prompting them to borrow and spend earlier, thus pushing up the risk-free real interest rate ('AI utopia scenario'); the other posits that improved productivity could boost government surplus prospects, thereby reducing interest rates. Regarding the phenomenon of declining US Treasury yields during AI model releases in 2023-24, UBS offers a cautious critique. Although previous studies attributed falling yields to the optimistic outlook on productivity enhancements from AI, UBS notes that this period also overlapped with CPI being below expectations and the Federal Reserve’s dovish pivot in December 2023. Therefore, the decline in yields cannot be solely attributed to AI itself. Expanding the analysis to include data up to 2025-26, UBS arrives at a clearer conclusion: when major AI laboratories (such as Google, OpenAI, Anthropic, xAI, Meta, DeepSeek, etc.) launch new state-of-the-art models, there is no meaningful decrease in the yield of the US 10-year bond. Additionally, by examining the spread between the 10-year US Treasury yield and swap rates—a metric commonly used to track changes in fiscal risk—there is no strong evidence that markets have significantly repriced US fiscal risk following AI releases. On macro implications, the report maintains that fiscal and monetary policy combinations remain key determinants of interest rates. Despite relatively high real interest rates in recent years, post-pandemic real personal income growth in the United States has been unusually stable. For AI's productivity dividend, European Central Bank official Schnabel and other policymakers emphasize that in the near term, AI is more likely to bring inflation rather than deflation. This is due to the large-scale investment demands of energy-intensive data centers, as well as emerging bottlenecks in specialized chips and skilled labor. Moreover, scholar Chad Jones notes that technology diffusion is typically incremental, constrained by the weakest part of the supply chain. He also warns that nations may underfund AI safety and estimates the willingness of the US to pay for safer AI could reach up to 100% of GDP.

Analysis framework

UBS primarily employed an event study methodology to analyze the actual market impact of AI model launches. Specifically, the research unfolded along the following lines: First, defining the event window. The report designated the dates when major AI laboratories (Google, OpenAI, Anthropic, xAI, Meta, DeepSeek) launched cutting-edge models as event days and selected a three-day time window before and after each event to observe market reactions. Second, selecting observation metrics and comparing against baselines. The report chose the change in the yield of the US 10-year treasury as the core observable and compared the yield movements on AI launch days with those around FOMC meetings, non-farm payroll (NFP) announcements, CPI releases, and other significant events to control for cyclical macro factors. Finally, introducing a proxy variable for fiscal risk to cross-validate. To test whether AI launches affected the market's perception of US fiscal conditions, the report used the spread between the 10-year treasury yield and the swap rate as an indicator of fiscal risk, assessing whether this spread underwent substantial revaluation following AI model launches. Through this analytical pathway—from theoretical hypotheses to empirical validation and excluding confounding factors—the report concluded that AI launches did not significantly alter interest rate or fiscal risk pricing.

Methodology notes

• Event Gaming and Behavioral FinanceEvent-driven analysis

  Event Study Method

  By examining asset price changes within a short time window surrounding specific events (like AI model launches), this approach determines whether such events have a significant impact on the market. In this study, the yield changes on AI launch days were contrasted with those observed around FOMC meetings, NFP releases, and CPI announcements over a three-day window to filter out macro-level noise.

• Fixed Income and Credit AnalysisSpread and Asset Quality

  Treasury Yield-Swap Spread as a Fiscal Risk Tracker

  The spread between government bonds and swaps is commonly used by the market to gauge sovereign fiscal or credit risk. When concerns about fiscal health rise, government bond yields relative to swap rates tend to increase, widening the spread. This indicator was utilized to verify if AI launches altered the market's pricing of fiscal risk.

• Industry/Industrial Analysis Framework

  Weakest Link Constriction Principle

  Economist Chad Jones proposed that technological diffusion is progressive, with production improvements often hindered by the most vulnerable part of the industrial chain. The report invokes this principle to illustrate that despite the promising outlook for AI technology, bottlenecks such as shortages in specialized chips or scarcity of skilled labor can limit its actual contribution to overall productivity.

Key data

• Permanent Additional Annual Productivity Growth10 basis points (bps)Scholarly estimates suggest this increment could mean a 70-basis-point reduction in the US 10-year interest rate
• Willingness of the US to Pay for Safer AIUp to 100% of GDPAn estimate by scholar Chad Jones of the upper limit for national willingness to invest in AI safety
• Sample Period for AI Model Release EventsFrom 2023 to April 2026Covering approximately 30 frontier model releases by major AI laboratories including Google, OpenAI, Anthropic, xAI, Meta, and DeepSeek

Impact & implications

The report argues that although some investors believe AI-driven productivity gains will quickly generate structural benefits across industries and the labor market, in the short term, AI is more likely to exert upward pressure on inflation rather than downward pressure. This implies that interest rates are unlikely to automatically decline due to AI factors, given the massive investment needs and potential supply chain bottlenecks. Furthermore, the incremental nature of AI technology diffusion means improvements to fiscal surpluses will not happen overnight. Fiscal and monetary policies will continue to dominate interest rate trends, rendering AI neither the central variable nor the immediate determinant of current interest rates and fiscal risk pricing.

Risks

• Risks associated with multi-asset investing include but are not limited to market risk, credit risk, interest rate risk, and foreign exchange risk; return correlations between different asset classes may deviate from historical patterns
• Geopolitical events and policy shocks could adversely affect valuations during periods of low liquidity and economic dislocation
• A phase of significant productivity gain could be disruptive, even if the final outcome is ultimately beneficial

What to watch

• The subsequent evolution of fiscal and monetary policy combinations and their impact on interest rates
• The development of potential bottlenecks related to energy-intensive data center investments, specialized chips, and skilled labor
• Policy developments and funding shifts concerning AI safety expenditures worldwide