BDCs' $115 billion software exposure flags Generative AI risk mispricing

The Expanding Footprint of Private Credit in Software

Private credit firms, particularly Business Development Companies (BDCs) in the United States, have significantly expanded their lending to the software sector over the past decade, accumulating substantial exposures that warrant scrutiny given emerging technological shifts.

As of late 2025, BDCs had extended approximately $115 billion in loans to software companies, which represents about one-fifth of their entire lending portfolio and over 80% of their technology-focused allocations. This growth has positioned information technology (IT) as the second-largest sector for BDC lending, reaching nearly $140 billion, following only business services.

This trend contrasts with the more stable and smaller share of IT in the broader US economy's output, revealing a concentrated investment focus by direct lenders.

The Bank for International Settlements (BIS) Bulletin No 128, published in July 2026 by Fernando Avalos, Giulio Cornelli, and Egemen Eren, highlights that this concentration in software has been driven by the sector’s historical characteristics: reliable recurring subscription revenues, high profit margins, and limited capital expenditure requirements, which made these firms attractive to direct lenders seeking stable cash flows.

However, the authors caution that these historical advantages are now subject to re-evaluation due to the rapid advancement of generative artificial intelligence, introducing a new layer of uncertainty for these significant loan portfolios.

Generative AI's Revenue Disruption Mechanism

The rise of generative artificial intelligence (AI) introduces a material shift in the revenue stability and competitive landscape for software firms, directly impacting the cash flows that service their loans.

Generative AI tools possess the capability to directly substitute existing software products, effectively diverting revenue from incumbent borrowers by offering comparable functionalities at a lower cost. For example, AI can perform tasks previously requiring specialised software, leading to direct displacement.

Beyond substitution, AI also intensifies competition by significantly reducing barriers to entry for new market participants. By lowering development costs and democratising advanced functionalities, AI enables new entrants to offer competitive products with thinner margins.

This dynamic forces established software firms to either integrate AI into their offerings, potentially at reduced profitability, or risk losing market share. Consequently, the cash flows underpinning the substantial loans to software companies have become subject to greater uncertainty than in prior periods.

While reported earnings may not yet fully reflect these pressures, the authors of the July 2026 BIS Bulletin emphasise that the long-term nature of these loans means that the impact of rapidly evolving AI capabilities could manifest before the loans reach maturity, challenging the initial assumptions of revenue predictability.

The Absence of AI Risk Pricing by Lenders

Despite the growing risk of generative AI disruption to software firm revenues, direct lenders have not adjusted their credit pricing to reflect this elevated uncertainty. Over recent quarters, credit spreads for software borrowers have not only failed to increase but have, in fact, narrowed, converging with spreads observed in other sectors.

By late 2025, software, other IT, and non-IT borrowers were paying broadly similar spreads on their loans, a trend particularly pronounced for newly issued loans, where spreads decreased more rapidly than for outstanding obligations.

Fernando Avalos, Giulio Cornelli, and Egemen Eren, writing in the July 2026 BIS Bulletin, note that this compression was even steeper for loans originated by non-listed Business Development Companies (BDCs), a segment subject to less market scrutiny.

While current performance metrics reveal no visible stress (less than 1% of software loans were behind on payments as of late 2025, a lower share than for other BDC portfolios), these backward-looking indicators may not fully capture emerging risks.

The authors suggest that a revenue shock takes time to translate into missed payments, and payment-in-kind arrangements can further delay the visibility of financial strain, creating a potential disconnect between observed credit quality and underlying risk.

Undifferentiated Equity Valuations for BDCs

Publicly listed Business Development Companies (BDCs) exhibit valuations that do not appear to differentiate based on their exposure to the software sector, despite a notable divergence in the performance of software equity indices.

The July 2026 BIS Bulletin by Fernando Avalos, Giulio Cornelli, and Egemen Eren notes that software-related stocks experienced a significant valuation premium over the broader technology sector following the pandemic, peaking by the close of 2021.

However, since December 2022, coinciding with the public release of ChatGPT, this premium has eroded, turning into a discount by the end of 2025. In contrast, BDCs with high concentrations of software loans trade at price/dividend ratios similar to those with lower software exposure, showing no discernible discount for software-specific concentration as of mid-2025.

This lack of differentiation in BDC valuations could suggest that investors primarily hold BDCs for their distributions, and as long as dividends remain stable and the distance to default does not critically diminish, slower-building credit risks may not immediately influence prices.

This presents a divergence in risk assessment between the equity markets for software firms themselves and the valuations of the financial intermediaries lending to them, raising questions about the thoroughness of risk integration in BDC pricing.

Concentrated Exposure and Eroding Loss Buffers

The private credit market's exposure to software firms exhibits concentrations that could amplify losses if generative AI disruption intensifies, particularly as loss absorption buffers have diminished. The July 2026 BIS Bulletin by Fernando Avalos, Giulio Cornelli, and Egemen Eren highlights two forms of concentration.

First, software borrowers increasingly rely on multiple lenders; approximately 60% of software lending now involves firms borrowing from seven or more BDCs, a significant increase from under 10% in 2015. While this diversifies funding for individual borrowers, it means that a software-specific credit event would simultaneously affect numerous BDC balance sheets.

Second, the lending side of the market is itself concentrated: the five largest BDCs account for about 37% of all software loans, and the top ten hold over half. This structure implies that a substantial credit event in the software sector would disproportionately impact a small number of major lenders.

Concurrently, the narrowing of credit spreads in recent years has reduced the interest income earned on software loans, leaving less buffer to absorb potential losses.

Lenders are therefore receiving less compensation for carrying a risk that, according to the authors, has demonstrably increased, making the system more susceptible to correlated losses and abrupt repricing if underlying credit quality deteriorates.

Structural Protections and Their Limits

While certain structural features of Business Development Companies (BDCs) are designed to limit wider financial spillovers, their effectiveness in mitigating generative AI-related credit risks remains subject to assessment.

Listed BDCs, which hold the majority of software exposure, are predominantly funded by equity investors who cannot redeem their capital, providing a stable funding base. These entities also have access to term funding and committed facilities for liquidity management.

Furthermore, BDC leverage is statutorily capped at low levels, considerably below that of traditional banks, and most loans are senior and secured, often including covenants that enable lenders to intervene early.

These provisions theoretically enhance recovery rates in the event of a default, as noted by Fernando Avalos, Giulio Cornelli, and Egemen Eren in the July 2026 BIS Bulletin. However, the aggregate quality of these structural protections and the extent of interconnections with other financial intermediaries are difficult to ascertain.

A particular vulnerability exists within the approximately one-quarter of software loans held by non-traded BDCs. These entities often feature semi-open structures with periodic, capped redemption options, typically 5% of assets under management quarterly.

Although intended to manage liquidity risk, these caps may offer less protection under stress, as cash buffers and loan repayments are often insufficient to fund repeated maximum redemptions.

Instances, such as a large non-traded BDC breaching its redemption cap in early 2026, demonstrate that funding fragility can emerge even with contractual provisions, potentially compounding any AI-triggered credit events into sustained liquidity stress.

The Broader Opaque Private Credit Landscape

The potential for disruption from generative AI extends beyond the visible segment of Business Development Companies (BDCs) into the broader, less transparent private credit ecosystem, posing a more systemic risk.

BDCs, which are subject to quarterly loan-by-loan reporting requirements by the US Securities and Exchange Commission (SEC), represent only a fraction of the direct lending market.

The July 2026 BIS Bulletin by Fernando Avalos, Giulio Cornelli, and Egemen Eren suggests that BDCs’ software exposure is broadly indicative of similar concentrations building in other direct lending segments, where disclosure is significantly thinner and interconnections with banks and other financial intermediaries are more challenging to trace.

These less observable parts of private credit operate under various legal exemptions and lighter regulatory oversight, meaning that vulnerabilities identified in BDCs could be amplified elsewhere without the same level of transparency or statutory protections.

Recent events have also raised questions about the quality of due diligence practices and loan covenants across the sector, further obscuring the true extent of risk. Stress tests indicate that a severe economic downturn could necessitate substantial asset sales across the entire private credit sector.

The authors conclude that it is likely through these opaque channels, rather than solely via BDCs, that disruption to software credit could propagate more widely throughout the financial system, potentially leaving the sector more susceptible than its current credit metrics suggest.

Implications for Asia's Financial Stability

The increasing exposure of private credit to software firms and the unpriced risks associated with generative artificial intelligence disruption present material implications for Asia's financial markets and investors.

While the primary data in the July 2026 BIS Bulletin focuses on US-based Business Development Companies (BDCs), the underlying dynamics of private credit growth and technological disruption are global.

Asian financial institutions, including banks, asset managers, and sovereign wealth funds, are significant participants in global private credit markets, either directly through their own lending operations or indirectly via investments in private credit funds.

Any widespread credit event or abrupt repricing within the software sector, particularly in opaque segments of private credit, could lead to capital reallocation or increased funding costs for Asian entities with similar exposures.

Furthermore, Asian economies are home to a rapidly expanding software industry and a burgeoning private credit market, which could mirror the trends observed in the US. Decision-makers should monitor local private credit market data for signs of concentrated software exposure and undifferentiated risk pricing, particularly in less regulated segments.

The specific datapoint to watch is the quarterly loan performance reports from major Asian private credit providers, alongside any regulatory pronouncements on sector-specific risk guidance, with the next significant update expected with Q3 2026 financial filings due in late October. This will indicate whether similar vulnerabilities are crystallising in Asia.