Is AI Inflationary or Deflationary?

Artificial intelligence promises a revolution in terms of productivity, leading to a wealthier, more efficient, and perhaps even more solvent economy. However, in the bond markets, this grand promise has yet to deliver the expected shock: announcements of increasingly efficient AI models have neither pushed down US yields nor visibly reduced the country's fiscal risk premium. In other words, while Wall Street admires the technology, the rates market is still waiting for proof.

One of the presumed benefits of AI is the boost to productivity. According to economic theory, such an improvement strengthens potential growth and helps alleviate the relative weight of public debt. Some economists even suggest that a small, permanent increase in productivity could sharply reduce long-term yields. In a strategy note published last week, UBS pointed out that a sustained annual productivity gain of 10bp could, in certain models, reduce the US 10-year yield by 70bp. The mechanics look elegant on paper: higher productivity leads to higher future revenues and less fiscal stress, resulting in lower rates. However, markets do not live within equations.

Since 2025, the successive releases of cutting-edge AI models (from OpenAI, Google, Anthropic, xAI, Meta, or DeepSeek) have not triggered any significant easing in US Treasury yields. This is noteworthy, as these launches are precisely the type of events likely to fuel optimism regarding future productivity. Yet, contrary to what might be expected in a scenario of technological euphoria, rates have not broken lower following these announcements. In fact, the opposite has largely been true, as shown in the following chart, which tracks the evolution of US 10-year (orange) and 30-year (blue) yields since the announcement of the first public ChatGPT in late 2022. The same observation applies to the spread between 10-year Treasury yields and swaps, often used as a barometer for fiscal risk: here too, there has been no spectacular rerating of fiscal risk following major AI announcements.

Evolution of US 10-year (orange) and 30-year (blue) yields since the announcement of the first public ChatGPT, late 2022

AI Has Not Yet Changed the Price of Time

Of course, AI is no economic mirage, although its impact is observed daily. Rather, it means that the market is still distinguishing between promise and proof. Productivity gains take time to diffuse: organizations, professions, infrastructure, software and habits all need to adapt. Above all, AI is not "free magic." It requires massive investment in data centers, energy, specialized chips and expertise. In the near term, these requirements may even be inflationary: increased demand for electricity, semiconductors and skilled labor in sectors that are already under strain. The technology intended to streamline the economy may therefore begin by creating new bottlenecks. This is exactly what appears to be happening at present.

The conclusion is less spectacular than model demonstrations, although more useful for investors: rates remain primarily driven by the "old world" (inflation, monetary policy, public deficits, debt issuance, fiscal credibility) before being reshaped by innovation. AI may profoundly change the economy, but it has not yet changed the price of time, i.e. interest rates. For now, markets are treating artificial intelligence as a call option on the future, rather than an immediate reduction in US sovereign risk. The great shift may yet come, especially as UBS points out that a "growing number of investors share the view that the tangible benefits of AI could manifest structurally, and sooner rather than later, in both the corporate world and labor markets."

The currently dominant scenario is one of inflationary AI during an initial phase, followed by a deflationary impact later. The duration of this first phase remains unknown. In the meantime, the bond market continues its role as the guardian of the temple.