Concerns about an AI-driven bubble are top of mind for investors. Capital spending has surged, with a lot more expected over the coming years. While AI-generated revenues remain modest, the expectation is that they’ll grow as companies embed AI into operations—boosting productivity and profits across sectors.
But how much of this optimism is already baked into stock prices? Our analysis suggests the S&P 500 is pricing in around 1.7 percentage points (%pts) of annual US productivity gains from AI— which would be an optimistic outcome. Our macro analysis suggests a more realistic baseline of 0.6%pts. That does not imply an equity correction, but it does signal more limited returns ahead, with annual index gains set to trail earnings growth.
Investment in AI infrastructure has emerged as a key driver of global economic growth
Investment in artificial intelligence (AI) has been a major driver of global growth this year, especially in America and Asia. Capital spending by the big five “hyperscalers” (Microsoft, Meta, Amazon, Oracle and Alphabet) has been so large that it shows up in macroeconomic aggregates. AI investment alone seems to have added roughly one percentage point to US investment growth in 2025, though a lot of the associated infrastructure—chips and servers in particular—was imported. The build-out will continue to prop up growth next year, though probably to a lesser degree.
Widespread AI adoption should lead to substantial and lasting impact on productivity and corporate earnings
The next step is the built-in phase, which will likely prove even more transformative and generate even larger revenue and earnings gains. Widespread AI adoption across sectors is set to have a substantial and lasting impact on companies’ earnings profiles, which are in aggregate closely correlated to GDP growth. Broad-based valuation gains in US equities since the launch of ChatGPT three years ago are a precursor of what is yet to come. The S&P 500 is trading close to the highest price-to-earnings ratio (PE) in more than 20 years and has added an estimated USD10tr in market capitalisation, a reflection of higher expected future earnings due to AI.
How much optimism is baked in US stock prices?
This raises a central question, which is the focus of this note: how does the re-pricing in equities and the implied AI benefit to future earnings stack against the economy-wide productivity growth assumptions this technology is expected to deliver?
Probably too much
Our work suggests that the market is likely too optimistic about future productivity gains and hence future earnings growth. Our first section shows that AI could add 0.6 percentage points (%pts), with a range of ±0.2%pts, to average annual productivity growth over the next ten years. Yet we also show in our second section that current S&P 500 valuations imply annual gains of 1.7%pts, well above our central estimate and those of most other studies. Such an outcome is not impossible, but it would require rapid and broad-based adoption across sectors, likely involving extensive deployment of AI-enhanced robotics, a scenario to which we attach a relatively low probability.
Equity gains are likely to be more limited than in recent years
Three implications follow. First, valuations appear stretched, limiting the scope for the kind of rapid equity gains seen in recent years. Second, investors are likely to become more selective in allocating capital. Even so, productivity gains from AI should become more evident over time. Sectors and firms best able to integrate the technology are likely to capture a larger share of the rewards.
What are the productivity gains we can expect from the use of AI?
The hype around AI, and in particular generative AI (GenAI), is well-founded and claims of a new industrial revolution are not unreasonable. Like earlier general-purpose technologies—from steam and railways to electricity, computers and the internet—its widespread adoption could lift productivity. But how large will the gains be, and when will they appear in the data? Predicting either is difficult. Research from the OECD, the IMF, and other economists provides a solid framework for estimating AI’s productivity gains and for evaluating the assumptions behind them, which this section examines.
In brief, overall productivity gains rise with AI exposure, adoption, and how broadly bene-fits are shared across sectors. Our central scenario—largely based on OECD modelling—assumes 40% workplace adoption within ten years (comparable to internet-era diffusion) and 35% of economic activity exposed to AI on average (with significant sectoral differences). Applying the OECD’s general equilibrium model, direct effects and supply-chain spillovers could lift productivity by 0.9%pts.
What’s priced in equity markets?
The US equity market has risen by about 70% since the launch of ChatGPT in November 2022, adding about USD25tr in market capitalisation. Roughly half of the gain comes from higher earnings, the other half is the result of rising valuations.
While some companies took the lead, the multiple expansion in the S&P 500 has by no means been limited to just a few names. In particular in the US, the re-rating happened quite broadly across sectors. In our view, this is a reflection of the market’s implicit assumption that not only a narrow set of companies will reap the direct benefit from AI sales, but that the entire growth trajectory of the economy will shift upwards as a result of AI adoption. Naturally, this begs the question: how much of an earnings and GDP growth acceleration is priced into equity markets? And how does this compare to economists’ estimates discussed above?
Our calculations illustrate that the equity market appears to factor in a highly disruptive AI scenario. The implied GDP growth impact of around 1.7%pts is on par with the most optimistic macro views from forecasters. This suggests that equity market gains in 2026 are likely to be more moderate than the returns which we have seen over the past three years. It does not necessarily imply, however, a substantial re-pricing of the market. If we were to see a seamless transition of build-out-driven growth to adoption and hence productivity-driven growth, the market can easily grow into current valuations. While gains will likely be more moderate, we still expect positive returns, though slightly below the pace of earnings growth.
Conclusion
AI infrastructure investment has been a key engine of global growth in 2025. As AI adoption accelerates and automates more tasks, productivity is expected to rise—potentially mirroring the surge seen with past general-purpose technologies. This should lift corporate earnings, which closely track GDP growth. AI’s transformative potential is undeniable, but our analysis suggests US equity markets are overestimating its economic impact. Indeed, while AI adoption is rising, real-world productivity and profit gains remain mixed. Most credible studies project AI-driven US productivity gains of 0.4– 0.9%pts annually; we forecast 0.6%pts (±0.2). Gains of 1%pts or more would require wide-spread AI-powered robotics—plausible, but not our base case. Current S&P 500 valuations appear to be pricing this optimistic scenario, with annual productivity gains of 1.7%pts, leaving markets exposed to disappointment. This doesn’t necessarily signal an equity market correction, but it does suggest more muted returns ahead, with annual price gains likely to lag behind earnings growth. Within equities, we prefer companies and sectors that are effectively adopting AI, rather than those that merely supply the infrastructure, with upside potential extending well beyond US tech stocks.
By Raphael Olszyna-Marzys, international economist at J. Safra Sarasin Sustainable Asset Management
