Prepared by Irene Ferrari, Sofia Gori, Johannes Groß, Fabian Schupp and Sigurd Steffensen
Published as part of the ECB Economic Bulletin, Issue 7/2026.
Artificial intelligence (AI) has been the dominant theme in global financial markets since late 2022, with equities in the technology sector outperforming substantially (Chart A). This outperformance has been especially striking in the United States, led by a small group of hyperscalers, and is evident among European firms that occupy upstream positions in the global AI infrastructure buildout. However, AI’s influence on markets extends beyond a narrow set of infrastructure providers and stretches further back in time. Many firms integrate AI into their businesses, either by developing AI-enhanced products or by adopting AI-enabled processes that improve productivity.[1] This box therefore introduces a new firm-level measure of AI exposure derived from AI-related discussions in earnings call transcripts. It uses this measure to examine how the AI exposure of European firms is reflected in their equity returns and corporate bond spreads.[2]
Chart A
Performance of the STOXX Europe 600 and S&P 500 and their technology subsectors
Firms are classified as AI-exposed when their earnings calls link AI to infrastructure provision, product development, or improvements in business processes and productivity.[3] These criteria are applied across STOXX Europe 600 constituents, identifying not only firms that supply AI technology but also firms across all sectors that embed AI in their products and operations. The approach also captures the evolving nature of AI: as the technology matures and diffuses across sectors, firms’ disclosures shift accordingly, allowing a measure grounded in contemporaneous statements to track this evolution. This is valuable given the reliance of analysis on timely identification of AI-related news. Accordingly, AI exposure is measured quarterly from 2017 onwards and treated as time-varying, allowing the analysis to control for firm heterogeneity and sector-specific time variation.[4]
The share of AI-exposed firms has risen markedly since 2022, with AI predominantly mentioned in earnings calls as a source of product innovation and growth (Chart B). 75% of earnings calls mentioning AI refer to product innovation or growth opportunities from an AI-enhanced product offering, while 50% discuss AI-related productivity gains, with some mentioning both. There is clear heterogeneity across sectors and over time: IT firms mentioned AI-related product innovation as far back as 2017, whereas firms in other sectors mainly started to refer to productivity improvements resulting from AI after the launch of ChatGPT in late 2022. This is consistent with the findings of the ECB’s Survey on the Access to Finance of Enterprises in the euro area in June and December 2025, which show more innovation-oriented cases of AI use in the IT sector and greater emphasis on cost reduction in manufacturing, construction, trade and other services.[5]
Chart B
Share of AI-exposed STOXX Europe 600 firms and breakdown by AI exposure type
b) Share by innovation and productivity exposure
Whether AI exposure raises or lowers firms’ financing costs is theoretically ambiguous. AI adoption may boost expected cash flows, lifting equity valuations and reducing perceived credit risk, but only if investors believe that firms are genuinely advancing, and not merely making unsubstantiated claims. On the debt side, AI adoption often requires substantial upfront capital investment concentrated in intangible assets with limited collateral value, potentially raising financing costs. Since these mechanisms operate simultaneously, the analysis examines whether AI exposure is systematically associated with different equity returns and bond financing costs than those faced by firms not exposed to AI.
Panel regression evidence points to an equity return premium of around 1.3 percentage points per quarter for AI-exposed firms (Chart C).[6] Quarterly stock returns are regressed on AI exposure flags, controlling for firm-level characteristics, earnings surprises, firm and sector-quarter fixed effects, and sector-specific time trends. This ensures that AI-exposed firms are compared with non-exposed firms in the same sector-quarter, netting out sector-wide movements.[7] The regression results align with descriptive patterns in the data: AI-exposed firms recorded higher median annual stock returns than non-exposed firms in most years (Chart C, panel a) and also tended to trade at higher price/earnings ratios (Chart C, panel b). This is consistent with a longer-duration equity profile, whereby much of the value of AI firms is tied to earnings that are expected to accrue far into the future.
Chart C
Stock returns and price/earnings ratios for AI-exposed and non‑AI‑exposed STOXX Europe 600 firms
|
a) Annual stock market returns |
b) Annual price/earnings ratios |
|---|---|
 |
 |
Non-financial corporate bond markets exhibit a broadly similar pattern to equity markets (Chart D). Panel regressions using quarterly option-adjusted spreads over 2019-26, with bond and issuer-specific controls, show that AI exposure corresponds to spreads that are around 5 basis points tighter – a modest but robust link between AI exposure and a lower market-implied cost of debt.[8] This again aligns with descriptive patterns in the data. AI-exposed issuers consistently exhibit lower spreads than non-exposed issuers, with both the median and interquartile range shifted downward across most years (Chart D, panel a). Exposed firms have also issued a larger share of bonds with maturities exceeding ten years (Chart D, panel b), whereas issuance by non-exposed firms is more concentrated at intermediate maturities. This combination points to more favourable financing conditions, since longer maturities would typically command higher, not lower, spreads.[9]
Chart D
Corporate bond spreads and bond maturity structure for AI-exposed and non‑AI‑exposed STOXX Europe 600 non-financial corporates
b) Maturity structure of new bond issuances
(left-hand scale: percentages, right-hand scale: percentages per annum)
The findings in this box suggest that AI exposure is positively reflected in equity and corporate bond pricing across the wider economy in Europe. AI’s association with lower market-based financing costs is more widely distributed than the concentrated equity rally in the IT sector suggests. However, this relationship rests on expectations about the technology’s returns. Should those expectations disappoint, valuations and spreads could reprice abruptly and broadly, extending well beyond the firms driving the AI investment cycle. Further analysis is warranted to establish the potential scale of this repricing and its effect on aggregate financing conditions in the euro area.[10]
References
Andersson, M., Breckenfelder, J., Corradin, S., Nikolov, K. and Viola, M.A. (2026), “The AI boom: rational enthusiasm or the next dot-com bubble?”, TheECB Blog, ECB, 17 August.
Acemoglu, D. and Restrepo, P. (2020), “Robots and Jobs: Evidence from US Labor Markets”, Journal of Political Economy, Vol. 128, No 6, April.
Babina, T., Fedyk, A., He, A. and Hodson, J. (2024), “Artificial intelligence, firm growth, and product innovation”, Journal of Financial Economics, Vol. 151, January.
Bergeaud, A., González-Torres Fernández, G., Labhard, V. and Sellner, R. (2025), “AI can boost productivity – if firms use it”, TheECB Blog, ECB, 28 March.
Calvino, F., Dernis, H., Samek, L. and Ughi, A. (2024), “A sectoral taxonomy of AI intensity”, OECD Artificial Intelligence Papers, No 30, OECD, December.
Ca’ Zorzi, M., Lopardo, G. and Manu, A.-S. (2025), “Verba volant, transcripta manent: what corporate earnings calls reveal about the AI stock rally”, Working Paper Series, No 3093, ECB, August.
Dias da Silva, A., Lebastard, L. and Sondermann D. (2026), “AI adoption and the productivity promise: what workers report”, TheECB Blog, ECB, 26 August.
Duquerroy, A., Furtuna, O., Rahmouni-Rousseau, I. and Vaz Cruz, L. (2026), “Big tech, big debt: when US tech giants tap the euro area bond market”, TheECB Blog, ECB, 31 August.
Ferrando, A., Lamboglia, S., Rariga, J. and Schmidt, M. (2026), “Adoption and investment in AI across the euro area: insights from harmonised firm-level data”, Occasional Paper Series, No 395, ECB, July.
Ferrari, I., Groß, J., Schupp, F. and Steffensen, S. (2026), “AI Exposure and the Costs of Market-based Financing in Europe: Evidence from Earnings Call Transcripts”, mimeo.
-
See, for example, Acemoglu and Restrepo (2020) and Babina et al. (2024).
-
Earnings calls have previously been used to measure the AI exposure of US firms (see Ca’ Zorzi et al., 2025).
-
This approach builds on Ferrari et al. (2026). Their prompt allows multi-label classification, anchors AI definitions to the time of each call, and provides a rationale for each flag, aiding interpretability and improving decision quality. Since firms may have an incentive to influence analysts’ perceptions when discussing their AI usage on earnings calls, the model guards against “AI washing” by excluding vague statements that are not explicitly linked to performance.
-
The measure correlates strongly with an established sector-level classification provided by Calvino et al. (2024), which combines indicators of AI human capital, AI innovation, barrier-adjusted exposure and AI use.
-
See Ferrando et al. (2026). Related ECB survey evidence also points to a rapid but uneven diffusion of AI, with around three-quarters of large euro area firms surveyed in 2024 reporting the use of AI in their business operations (Bergeaud et al., 2025). Complementary evidence from the ECB Consumer Expectations Survey shows that the share of workers reporting AI use at work roughly doubled between 2024 and 2026, with substantial heterogeneity across occupations and worker characteristics (Dias da Silva et al., 2026).
-
Since many firms report more than one type of AI exposure, the categories are not mutually exclusive and the analysis therefore focuses on firms’ overall exposure.
-
Importantly, the premium does not merely reflect the broader tech rally. Excluding the IT sector from the sample leaves the estimated premium similar in magnitude and significance, confirming that the return differential is present across the wider corporate universe. The premium is identified from within-firm variation, comparing the same firm in quarters with and without concrete AI disclosure, net of earnings surprises released at earnings calls and of lagged firm fundamentals and momentum.
-
In contrast to the equity findings, none of the three exposure types is individually associated with tighter spreads when entered separately. This suggests that bond markets price overall AI engagement rather than differentiating across specific forms, which is consistent with many firms reporting more than one exposure type.
-
Relatedly, US hyperscalers funding the buildout of AI infrastructure through European corporate debt markets are also found to issue bonds with longer maturities. For these firms, however, AI-related debt issuance is associated with higher, not lower, credit spreads relative to peers, as investors appear to demand a risk premium for the uncertainty surrounding firms’ AI investments. See Duquerroy et al. (2026).
-
For further information on the valuation risk associated with AI and the potential impact of a correction in US equities on euro area financial conditions, see Andersson et al. (2026).
