AI Advances Push Down Long-Term Bond Yields as Consumption Growth Expectations Dim

The new working paper “Do Markets Believe in Transformative AI?” by Isaiah Andrews and Maryam Farboodi examines how U.S. bond yields respond to releases of major AI models. Using 15 release dates from five leading labs (OpenAI, Anthropic, Google DeepMind, xAI, and DeepSeek) over the period January 2023–December 2024, the authors find that long-term Treasury, TIPS, and corporate bond yields decline around those releases and stay down for weeks. The magnitude of the decline exceeds 10 basis points on average, sustained over roughly 15 trading days [1][2].

From an asset-pricing perspective, long-term yield falls of this size suggest that bond investors are revising down expected consumption growth or lowering perceived probabilities of extreme positive scenarios (e.g., post-scarcity economy) — in other words, the bond market seems skeptical that AI’s transformative potential will translate into broadly distributed gains in standards of living [1]. The lack of movement in short-term yields supports this: markets don’t expect effects immediately but see them materializing in the medium to long term [2].

At the same time, sentiment measures of AI progress show a more positive reaction. On Metaculus, forecasters updated AI progress timelines more optimistically when model releases occur, expecting AGI-like developments sooner [2]. However, this seesaw doesn’t eliminate the yield response: greater optimism about AI progress comes alongside bond markets betting small shifts in consumption growth or risk, but not larger macro growth boosts [2].

Strategically, this divergence between tech optimism and bond market caution has multiple implications. For firms investing in AI infrastructure, being aware that bond markets are pricing in risk or limited returns to widespread consumption growth means greater emphasis might be placed on defensible business models over speculative scale. For macroprudential and monetary policy, the findings suggest that markets foresee potential labor market disruptions, inequality, or uneven income gains, which could dampen aggregate demand and complicate inflation forecasts. For asset allocators, this suggests long-duration assets could benefit from these cautious expectations if inflation remains subdued.

Open questions remain. Do specific types of AI releases (open-source vs proprietary, small vs large model, safety-oriented vs commercial) trigger different market responses? How robust are these yield reactions when controlling for concurrent macroeconomic surprises? And, importantly, do equities respond differently, or are they pricing in growth upside that bond markets don’t believe in?