AI openness shapes inflation and growth expectations

Economists have long argued that expectations about future productivity growth are reflected in long-term interest rates. If investors believe AI will raise long-run growth, they should demand higher yields to compensate for stronger future demand and potential inflationary pressure. Conversely, if AI is expected to concentrate gains without broad productivity spillovers, yields may fall.

Financial markets are beginning to treat artificial intelligence releases not as isolated technology announcements but as signals with macroeconomic consequences. The growing divide between open-weight and proprietary AI models is now shaping expectations about economic growth, inflation, and risk in ways that extend far beyond the technology sector.

That shift is examined in the research paper Market Beliefs about Open vs. Closed AI, published as a 2025 working paper by economist Daniel Björkegren of Columbia University. The study analyses how financial markets react differently to open-weight and closed AI model releases, offering new evidence that licensing choices in AI development are already influencing long-term interest rates, inflation expectations, and investor sentiment.

Markets read AI openness as an economic signal

The study is built around a simple but underexplored question: do financial markets respond differently when AI models are released openly versus when they are kept proprietary? To answer this, Björkegren conducts an event-study analysis covering 47 major AI model releases between late 2022 and 2025, split almost evenly between open-weight and closed models. The releases span leading developers across the United States, Europe, and China and include many of the most influential large language models introduced during the recent AI surge.

The findings reveal a striking divergence. Long-term bond yields consistently move in opposite directions depending on whether an AI model is released openly or kept closed. Closed-model releases are followed by declines in long-term yields, while open-model releases are followed by increases. The pattern is strongest at longer maturities, particularly for 20-year and 30-year U.S. Treasury bonds, which are most sensitive to expectations about long-run growth, inflation, and risk.

This divergence suggests that markets interpret open and closed AI releases as fundamentally different economic events. Closed releases appear to signal contained diffusion of AI benefits, with gains accruing primarily to a small number of firms. That interpretation aligns with lower expected growth volatility and downward pressure on long-term yields. Open releases, by contrast, appear to signal broader diffusion of AI capabilities, faster adoption across the economy, and potentially higher growth or greater uncertainty, pushing yields upward.

Importantly, the study finds that these reactions are not driven by differences in technical quality. When controlling for model performance, distance from the technological frontier, or perceived progress toward artificial general intelligence, the licensing structure remains the dominant factor explaining yield movements. This indicates that markets are responding less to what the models can do and more to who can use them and how widely their benefits might spread.

The same pattern extends beyond government bonds. Corporate bond yields, inflation-linked securities, and currency markets show aligned movements. Open-model releases tend to coincide with higher inflation expectations and a stronger U.S. dollar, while closed releases align with lower inflation expectations. Together, these responses point to a broader reassessment of macroeconomic conditions triggered by AI openness rather than isolated sector-specific reactions.

Open versus closed AI reshapes growth and risk expectations

Economists have long argued that expectations about future productivity growth are reflected in long-term interest rates. If investors believe AI will raise long-run growth, they should demand higher yields to compensate for stronger future demand and potential inflationary pressure. Conversely, if AI is expected to concentrate gains without broad productivity spillovers, yields may fall.

The findings suggest that markets associate open AI with stronger or more uncertain growth trajectories. Open-weight models allow a wider range of firms, researchers, and developers to build on the technology, potentially accelerating diffusion and experimentation. That openness can increase investment demand, raise uncertainty about competitive dynamics, and amplify both upside and downside risks. These effects would be consistent with rising long-term yields and higher inflation expectations.

Closed models send a different signal. By keeping weights proprietary, developers retain control over access, pricing, and deployment. Markets may interpret this as limiting diffusion and reducing competitive disruption. While this could constrain broader productivity gains, it also reduces uncertainty and the likelihood of rapid, economy-wide transformation. That interpretation aligns with falling long-term yields and more subdued inflation expectations following closed releases.

The paper carefully avoids claiming a single definitive mechanism. Instead, it outlines several plausible channels through which openness could affect market beliefs. One possibility is that open models redistribute the gains from AI more broadly across firms, increasing aggregate investment and demand for capital. Another is that openness raises the probability of extreme outcomes, both positive and negative, including faster growth or higher systemic risk. Closed models, in contrast, may signal a narrower distribution of gains and more predictable outcomes.

The study also tests whether markets react differently depending on where open models originate. Given concerns about geopolitical competition, particularly between U.S. and Chinese AI developers, the paper examines whether open releases from Chinese labs trigger distinct responses. It finds no consistent evidence that country of origin explains the yield divergence. Licensing structure, rather than national origin, remains the primary driver of market reactions.

Equity market responses are more mixed, particularly for large technology firms. In some periods, open releases coincide with weaker equity performance for AI-intensive firms, possibly reflecting expectations of increased competition and reduced monopoly rents. However, these effects are less stable than bond market responses and vary across time and firms. The bond market evidence, by contrast, is consistent and robust across specifications.

Why AI licensing now matters for policymakers and investors

AI releases are no longer viewed as niche technology events. Markets are treating them as macro-relevant signals that shape expectations about growth, inflation, and risk over decades. This has direct relevance for policymakers, regulators, and central banks as they assess the broader economic impact of AI development.

For policymakers debating AI openness, the findings highlight a trade-off that extends beyond innovation and safety. Decisions about whether to encourage open or closed AI models may influence investment behavior, inflation dynamics, and long-term interest rates. Open AI could support broader productivity growth but also increase uncertainty and pressure on capital markets. Closed AI could stabilize expectations while concentrating economic power and limiting diffusion.

The study also adds nuance to debates about AI regulation. Much of the current policy discussion focuses on safety, misuse, and governance. Björkegren’s analysis suggests that markets are already pricing in the economic consequences of licensing choices. Regulatory decisions that tilt the balance toward openness or proprietary control could therefore have unintended macroeconomic effects.

For investors, the findings underscore the importance of tracking AI release strategies, not just technical breakthroughs. The licensing model of a new AI system may provide as much information about future economic conditions as traditional macroeconomic indicators. As AI becomes more central to productivity, its development pathway is increasingly relevant to asset pricing across markets.

The paper is careful to acknowledge its limitations. The sample of AI releases remains relatively small, and the AI landscape is evolving rapidly. Market responses observed today may change as investors gain more experience with AI diffusion and as regulatory frameworks mature. The author emphasizes that these findings should be viewed as early evidence rather than definitive conclusions.

Nevertheless, the results suggest that financial markets are already forming beliefs about how AI will reshape the economy, and those beliefs differ sharply depending on whether AI is open or closed. As AI continues to advance, these expectations are likely to become more pronounced, reinforcing the link between technology policy and macroeconomic outcomes.