Artificial inbreeding: BIS paper warns AI investment loops may amplify financial risks

A Bank for International Settlements (BIS) analysis examines the rapid expansion of “circular” investment relationships in the artificial intelligence sector: arrangements in which two AI companies have both a financing relationship and a commercial supplier-customer relationship. With major AI companies planning hundreds of billions of dollars in infrastructure spending, these connections have become potentially significant beyond the individual firms involved.

Using data covering 1,246 AI companies across compute, infrastructure, data tools, models and applications, the authors identify 972 intra-AI investment relationships between 2021 and 2025. By deal value, 28.7% of the investments made by AI firms targeted another AI company, while 55.2% of investment flowing into AI companies came from other AI firms.

Circularity is particularly important when measured by the amount of capital involved. While only 16.1% of AI-to-AI deals by number involved firms that also had a commercial relationship, those transactions represented 46.4% of disclosed AI-to-AI deal value. In 64% of circular relationships, the investing company also supplied products or services to the company receiving its investment.

The BIS stresses that these arrangements can have a clear economic rationale. Suppliers may finance customers because they possess better information about their business prospects and because financing can stimulate demand for their own products. This is particularly relevant for AI model developers, which require very large amounts of capital and depend heavily on cloud computing, chips and other costly inputs before generating comparable revenues.

The relationship can also work in the opposite direction. AI companies may finance suppliers to secure access to scarce or strategically important inputs such as memory chips. Equity stakes and financing can also help align incentives where infrastructure, chips, models or data centres are highly customised to a particular commercial relationship, effectively providing some of the benefits of vertical integration without a full acquisition.

These conditions are unusually pronounced in AI. The sector combines information asymmetries, uncertain prospects for young firms, highly specialised inputs, a limited number of critical suppliers and exceptionally large financing requirements. The BIS finds that 73% of circular investment relationships originate from companies in the compute or infrastructure layers, underlining the strategic importance of the upstream AI supply chain.

The report nevertheless identifies potentially important macroeconomic risks. Supplier financing can make reported demand partly dependent on the supplier’s own investment decisions: a company may finance a customer that then uses the money to purchase its products. This makes it more difficult for investors, lenders and supervisors to distinguish genuinely independent end-user demand from demand supported by intra-industry financing. The BIS draws a parallel with vendor financing during the telecommunications boom of the late 1990s.

Circular relationships can also create double exposure to the same counterparty. A company that both invests in and supplies another firm risks simultaneously losing investment value and future sales if that customer comes under pressure. Because many such relationships involve large upstream companies, shocks could spread through financial and commercial channels at the same time. Private credit and special-purpose vehicles used to finance AI infrastructure could further add hidden leverage and interconnected exposures.

Finally, the BIS highlights a transparency problem. Many AI firms are private, while even publicly disclosed transactions can combine equity investments, long-term purchase commitments and guarantees. Some commitments, such as residual value guarantees for chips or data-centre equipment, remain off balance sheet until stressed conditions materialise. This means headline transaction values may provide an incomplete picture and that risks spanning firms, sectors and jurisdictions can be difficult for any single supervisor to monitor.

Source – Bank for International Settlement