Renting Intelligence: Vendor Concentration Risk and the Pricing of AI Dependency
This paper finds that equity markets do not currently price the risk of AI vendor concentration because mandatory disclosure rules are absent, leading to inconsistent and often missing information in public filings that prevents investors from distinguishing between firms that rent versus those that own their AI models.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the modern economy, companies are increasingly building their products on top of artificial intelligence. To do this, a business faces a fundamental choice. It can either license the intelligence from a large technology provider, paying a fee every time the system is used, or it can build and run its own systems from scratch, investing heavily in hardware and training data upfront. This decision is not just a technical one; it is a financial gamble. When a company rents the intelligence, it relies on a price set by someone else, a price that could change at any moment. When a company builds its own, it has already paid the cost and owns the capacity. Financial theory suggests that relying on an outside provider for a critical input should make a company riskier, much like a business that depends on a single giant customer is considered more fragile than one with many small clients. Investors, who constantly weigh risk against reward, should theoretically charge these "renting" companies more to lend them money, or demand higher returns on their stock, because their future costs are less certain.
A researcher at the Peres Academic Center set out to test whether the stock market actually sees this difference. The study examined thousands of annual reports filed by American companies between 2023 and 2026. The goal was to sort these companies into two groups: those that admitted to renting their artificial intelligence models from others, and those that claimed to build and host them themselves. The researcher then compared the stock behavior of these two groups against companies that did not use the technology at all. The expectation was that the renting companies would show signs of higher risk, such as more volatile stock prices or higher borrowing costs, compared to the builders.
The results, however, were not what the theory predicted. The study found that the stock market could not tell the difference between the companies that rented and the companies that built. Both groups carried similar levels of risk, and neither group was priced differently from the other. The only clear distinction the market made was between companies that used the technology at all and those that did not. Companies that used artificial intelligence, regardless of how they obtained it, appeared riskier than those that did not. But the specific choice of renting versus building did not move the needle.
The reason for this silence in the data is not that the risk does not exist, but that the companies are not saying enough about it. The researcher discovered that most companies writing about artificial intelligence in their annual reports stop just short of the truth. They might claim to have a "proprietary platform" or "in-house technology," but they rarely specify whether the models running underneath are actually their own or if they are licensed from a third party. Because the companies do not explicitly state where their models come from, the researcher had to guess based on vague language. When two independent readers tried to classify the same reports based on these vague descriptions, they agreed on the answer only about half the time.
This lack of clarity created a major blind spot. The study showed that the "renting" group and the "building" group were so poorly defined in the reports that the market could not possibly distinguish between them. It is like trying to measure the difference between two types of fuel when the labels on the tanks are smudged and unreadable. The data was simply too noisy to reveal the risk. The researcher calculated that even if a real difference existed, the study's design was not sensitive enough to find it unless that difference was enormous—roughly a quarter of the total risk variation seen in the market.
The most telling finding came from looking at the companies that mentioned artificial intelligence but refused to say where their models came from. These companies, who spoke about the technology without revealing its source, actually carried higher risk than companies that never mentioned the technology at all. Their stocks were more volatile, and investors demanded a higher return for holding them. This suggests that the silence itself is a signal of risk. When a company uses a powerful technology but hides its origins, the market senses something is uncertain, even if it cannot pinpoint exactly what.
The study concludes that the current way companies report their use of artificial intelligence is insufficient for investors to make informed decisions. The market is not ignoring the risk of renting; it simply cannot see it. The reports are too vague, and the language is too inconsistent to allow for a clear measurement. The researcher suggests that a simple change in reporting rules could fix this. If regulators required companies to state clearly whether their models are rented or built, and to name the provider if the dependency is significant, the data would become clear. Until that happens, the specific risk of renting intelligence remains invisible, hidden behind a sentence that most companies choose not to write. The market knows the technology is risky, but it does not yet know which companies are holding the bag.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.