Davidson Kempner Capital Management, one of the largest alternative credit managers with over $45 billion in assets under management, has warned investors that AI poses a material risk to recovery values in private debt portfolios with significant software company exposure, arguing that the collateral assumptions, renewal rate durability, and exit multiple expectations that justified lending terms across a generation of private credit software deals may be systematically overestimated as AI compresses software category margins, enables enterprise buyers to reduce seat counts, and creates viable alternatives to categories that private credit treated as recurring revenue with structural moats.
The private credit market's exposure to software company debt is larger and more concentrated than most discussions of AI disruption risk acknowledge. The leveraged buyout wave in enterprise software from 2018 through 2022 produced hundreds of transactions where private equity sponsors acquired software companies at 10 to 20 times revenue, financed the acquisition with a combination of sponsor equity and private credit debt, and projected the debt service from the target's recurring subscription revenue. The underwriting logic was well-suited to the software business model of that era: subscription revenue was sticky, net revenue retention above 100% meant the revenue base grew without new customer acquisition, and the enterprise software category had demonstrated through multiple economic cycles that customers renewed their contracts even in downturns because switching costs and integration depth made cancellation expensive. Private credit lenders pricing that revenue stream as collateral applied haircuts that reflected those assumptions, and the resulting loan terms reflected a belief that software revenue was more durable than revenue from most other sectors.
Davidson Kempner's argument is that those underwriting assumptions were calibrated for a software market that is structurally different from the one that now exists and is becoming more different each quarter. AI is changing the software market in three specific ways that affect private credit collateral quality simultaneously. The first is margin compression in software categories where AI-native competitors can deliver equivalent or superior functionality at lower prices because their development and infrastructure costs are materially lower than those of established vendors built on legacy codebases. A company that borrowed at eight times EBITDA on the assumption that its gross margins would remain in the high 70s is in a structurally different financial position if AI-native competition compresses those margins to the mid-60s over the debt's life. The second is seat count reduction. Enterprise software pricing has historically been seat-based, and the number of seats a company licenses to a vendor has been relatively stable because headcount was the limiting factor for software users. When AI agents replace human users in specific workflows, the enterprise's seat count for that software category declines even if the overall workflow value is maintained. A company that underwrite a seat-based software revenue stream is holding collateral whose pricing mechanism is being eliminated rather than grown. The third is demand substitution. Categories like contract management, customer success automation, content creation, coding productivity, and business intelligence reporting all have established SaaS vendors with hundreds of millions of dollars in private credit financing, and all of them face AI-native alternatives that replicate their core functionality with better natural language interfaces and significantly lower switching costs than the integrated enterprise software of the prior era.
The specific software categories Davidson Kempner and other credit managers examining this risk most closely are those where AI substitution is most direct and most near-term. Content and copywriting tools face Generative AI that eliminates the category entirely for many enterprise use cases. Customer support and contact center software faces AI voice and chat agents that reduce seat count and software spend simultaneously. Legal research, document review, and contract analysis tools face AI models that replicate their output without the licensing and integration costs. HR workflow software that processes approvals, routes requests, and generates standard communications faces automation that reduces the number of interactions requiring the software to process. The common thread is that these are categories where the value delivered was historically tied to human productivity augmentation at a per-seat pricing level, and where AI can now deliver the productivity outcome without the human, which eliminates the per-seat revenue model entirely rather than simply making it more efficient.
The distress opportunity framing is the angle that makes this relevant for AI-native startup founders rather than just a credit market concern. Davidson Kempner's warning is simultaneously a prediction that certain software assets will trade at distressed valuations if their revenue assumptions prove incorrect and an implicit identification of which software categories are most likely to become acquisition targets at compressed multiples. An AI-native company that has built a superior version of a workflow automation, document processing, or customer service tooling product can access a generation of legacy software companies, their customer relationships, their enterprise contracts, and their data assets, at prices that reflect the private credit market's repricing of their collateral quality. The distress cycle that Davidson Kempner is describing, if it materialises on the timeline that current AI adoption rates suggest, would produce a set of corporate sales, covenant violations, and forced restructurings that create acquisition opportunities for AI-native companies with strong balance sheets and strategic interest in specific customer bases or workflow categories.
The financing implication for software startups that are not AI-native is the other side of the same dynamic. Private credit funds that have historically been the primary non-dilutive capital source for profitable SaaS businesses above Series B are repricing their lending criteria for software categories they now consider AI-exposed. A company with strong historical NRR and a defensible customer base in a category Davidson Kempner's analysts have flagged may find that the debt terms available to it in 2026 are more expensive, more restrictive, or simply unavailable compared to what a comparable company could have borrowed at in 2022. That repricing is not a catastrophe for companies with strong balance sheets and AI integration roadmaps. It is a significant constraint for companies that planned to use private credit to finance growth or acquisition activity and now face lenders who have updated their assumptions about the durability of software revenue in the categories Davidson Kempner has identified as exposed. The AI boom is feeding back into software financing exactly as Davidson Kempner has described, and the companies that prepared for that repricing by demonstrating AI integration depth, expanding to outcome-based or usage-based pricing, and building defensible data network effects within their category are in substantially better position than those that continued to operate on seat-based revenue models without a visible AI strategy.
Also read: Developers Are Reporting Claude Opus 4.7 Coding Regressions and the Complaint Pattern Points to a Deeper Problem Than One Model Version • vLLM's Merged TurboQuant Fix for Qwen 3.5 Is a Quiet Infrastructure Update That Changes the Serving Economics for a Model Tier Founders Were Already Watching • Jensen Huang Says AI Is Creating an Enormous Number of Jobs and He Is Right About the Chip Ecosystem and Wrong to Leave the Rest Unexplained
