“We will not use the unsecured debt market to fund our business at current yields, nor do we need to,” CoreWeave’s chief executive, Michael Intrator, told analysts on the company’s earnings call on 10 November. According to TechCrunch’s report of the call, yields on some of the company’s bonds had climbed above 15 percent in October. CoreWeave rents out computing power for AI. Wall Street took its results well, and its bond investors, TechCrunch says, were much less sanguine.
Two days later, the European Central Bank published a short study that describes a very different corner of the market. In a box in its Economic Bulletin, five ECB economists (Irene Ferrari, Sofia Gori, Johannes Groß, Fabian Schupp and Sigurd Steffensen) report that European companies that talk concretely about AI are rewarded twice, in the price of their shares and in the price of their debt. The gap between the two pictures is the story.
What the ECB measured
The authors read the earnings-call transcripts of the companies in the STOXX Europe 600 index, quarter by quarter, from the first quarter of 2017 to the second quarter of 2026. A company counts as AI-exposed in a quarter when its management links AI to infrastructure, to product development, or to better business processes and productivity. The box says the classification was made by a language model with a prompt built to exclude vague statements that are not tied to performance, a guard against what the authors call “AI washing”. The prompt and the method are described in a working paper by Ferrari and colleagues that is still a draft (mimeo), so the measure itself has not been published for outsiders to test.
On the calls that mention AI, 75 percent refer to product innovation or growth from an AI-enhanced product, and 50 percent discuss productivity gains; some mention both. The mix depends on the sector. Technology firms were talking about AI products as early as 2017, while firms in other sectors began to speak of productivity gains only after ChatGPT launched in late 2022.
Two premiums
The first result concerns shares. In a panel regression that compares AI-exposed firms with unexposed firms in the same sector and quarter, and controls for earnings surprises and firm characteristics, exposed firms earn about 1.3 percentage points more per quarter. The authors say the premium is not simply the technology rally: removing the information technology sector leaves it similar in size and significance, and they identify it by comparing the same firm in quarters with and without concrete AI disclosure. Exposed firms also trade at higher price-to-earnings ratios, which the box reads as a sign that much of their value lies in earnings expected far in the future.
The second result concerns bonds. Using quarterly option-adjusted spreads for euro bonds of non-financial companies from 2019 to 2026, the authors find that exposure goes with spreads about 5 basis points tighter, which they call “a modest but robust link”. Exposed firms also issue a larger share of bonds that mature in more than ten years, which would normally cost more, not less. Here the box adds a detail: none of the three exposure types (product, productivity, infrastructure) predicts tighter spreads on its own. Bond buyers, it concludes, price overall engagement with AI rather than any particular use.
Both results are associations. The box says outright that it is theoretically unclear whether AI should lower or raise a firm’s financing costs. Adopting AI may lift expected cash flows, but it also tends to demand heavy spending on intangible assets that make poor collateral. The data show which effect has dominated in the market so far, not that AI caused it.
The exception inside the box
A footnote in the same text pulls the other way. United States hyperscalers that finance their AI buildout in the European corporate bond market do issue longer-dated debt, but their spreads are higher than their peers’, not lower. The footnote draws on earlier ECB analysis and says investors appear to demand a risk premium for the uncertainty around these firms’ AI investment.
That is the CoreWeave picture at a larger scale. TechCrunch reports that CoreWeave expects about $9 billion in customer prepayments and deposits over the next three months and is leaning on short-term contracts that let it charge higher prices, though lenders, the report notes, give more weight to longer contracts because they value predictable cash flow. Where a company says AI will improve its products, markets have so far paid for it. Where a company must borrow heavily to build the machines, they charge.
What the premium rests on
The authors end with a warning. The relationship, they write, “rests on expectations about the technology’s returns”, and if those disappoint, “valuations and spreads could reprice abruptly and broadly”, reaching well beyond the firms that drive the investment cycle. They call for further work on the potential scale of that repricing.
The evidence of returns on the ground is still thin. In its November Monetary Policy Report, the Bank of England says firms surveyed estimate that AI has raised their productivity by around 0.5 percent over the past three years, which it translates into 0.1 to 0.2 percentage points of annual productivity growth since early 2024. The same firms report close to zero effect on total employment. The Bank finds that AI’s contribution to UK growth so far comes from building and implementing the technology, not from gains in the firms that use it. Firms expect much more, about 0.9 percentage points a year over the next three years; the Bank treats that as an upside risk with large uncertainty.
So the ECB’s two premiums measure what investors expect a company’s AI talk to become. A company whose managers can name a product or a saving in front of analysts is, for now, a cheaper borrower and a more expensive stock. The box does not say whether the talk comes true. It says what happens to those prices if it does not.
References
- Bank of England (2026). Monetary Policy Report, November 2026. bankofengland.co.uk
- Bellan, R. (2026). CoreWeave turns to customer prepayments, shorter contracts to fund AI growth. TechCrunch, 11 November. techcrunch.com
- Ferrari, I., Gori, S., Groß, J., Schupp, F., & Steffensen, S. (2026). Artificial intelligence and European firms: the cost of market-based financing. ECB Economic Bulletin, Issue 7/2026. ecb.europa.eu