FitchRatings AI Stress Test applies a common sector scoring framework to describe how adverse artificial intelligence (AI)- related developments could affect credit profiles.
NurPhoto via Getty Images
Fitch Ratings finds that AI is not a broad threat to credit quality. Yet, it flags sharp pockets of risk in outsourced services, media production, software and AI infrastructure, and in the private credit vehicles that lend to them.
Fitch Ratings recently published its Artificial Intelligence Stress Test, assessing how adverse AI developments could affect credit profiles over five years. The headline is reassuring: sectors representing the vast majority of rated debt show resilience. However, the detail is less comfortable. A handful of corporate sectors score between 60 and 80, and the financial institutions that lend to or invest in them carry that risk.
- Most sectors hold up. Only a few corporate sectors score 60 or higher. Business process outsourcing (BPO) and media outsourced production services top the list at 80.
- Banks and private credit score alike but fail differently. Retail banks and business development companies (BDCs) both score 40. For banks, the main threat is a slow shift in the business model. For BDCs, it is loan losses, especially on software borrowers.
- Nothing has broken yet. Fitch reports no AI-driven rating actions for BDCs, fund finance, retail banks, corporate and investment banks, or data centers to date.
How Fitch’s Artificial Intelligence Stress Test works
This first phase scores a representative issuer in each of 107 granular sub-sectors across Corporates, Infrastructure, Financial Institutions and Structured Finance. Public Finance and Sovereigns are excluded. Fitch tests three adverse scenarios, each calibrated to roughly a 10% to 20% likelihood over five years. They are not mutually exclusive:
- Disruption: AI is deployed quickly, including on tasks once thought too complex. Barriers to entry and switching costs erode, which brings new competition, lower profitability and tighter access to capital.
- Over-investment: Investors lose confidence in AI returns. Hyperscaler capex falls sharply, AI-native firms hit financial difficulty, circular financing arrangements unwind, and AI-specific data centers and equipment lose value.
- Asset impairment: Weaker performance at portfolio companies or underlying assets pushes up leverage and causes losses at financial institutions.
Each sector gets a score from 0 to 100. Fitch stresses that the rating impacts are illustrative and that a score of 40 or more signals the potential for negative rating action for a typical issuer.
Illustrative Rating Scores
MRV
Where The Stress is Highest
The highest scores sit in labor-intensive services, media and software, plus the infrastructure built for AI.
A high score is not a forecast of imminent downgrades. In BPO, rating actions have been muted relative to the score of 80, because many incumbents benefit from high switching costs and deliver services through AI themselves. Issuers with hard-to-replicate services can score below the sector.
Fitch’s AI Stress Scores
MRV
AI Training Data Centers Versus Cloud Data Centers
The split between the two data center categories shows how Fitch analysts think about over-investment. Cloud data centers in strong markets have a use case beyond AI and typically carry fully contracted leases of 7 to 15 years with highly rated hyperscalers. AI training facilities tend to be remote and used mainly for AI, and Fitch does not rate speculative-build AI data centers. The more exposed group is corporate-rated “neoclouds” that own the chips and sell compute directly. For them, stress would arrive through lower utilization, weaker pricing and customer concentration.
Financial Institutions: The Same Score Can Mean a Different Risk
Retail banks, private banks, BDCs, fund finance and investment management organizations all score 40. The difference lies in how stress reaches each one.
Fitch’s Financial Institution Scores
MRV
Banks: A Slow Squeeze, Not a Credit Event
Fitch expects direct consequences for banks to be neither widespread nor severe enough to drive sector-wide rating actions over five years. The pressure points are competitive: fee and margin compression in standardized products, and smaller banks that cannot match larger rivals’ AI spending. Corporate and investment banks score only 20 because their franchises rest on client relationships and regulatory barriers. Their main exposure is indirect, through private credit and collateral-dependent financing businesses.
BDCs: Software Is The Exposure
Fitch’s BDC score reflects asset risk. Software is a top industry exposure in many BDC portfolios, and Fitch estimates it at about 20% of portfolios at fair value, with some reported figures understated. Software valuations have already fallen and pressured net asset values. The adverse scenario assumes meaningful software write-offs.
The cushion looks adequate for now. Fitch expects most rated BDCs to keep meeting asset-coverage requirements even if every software investment were written down by 50%. Software loan maturities pick up meaningfully from 2028, and the downside outliers are BDCs with outsized software exposure or elevated starting leverage.
Fund Finance And Asset Managers
In fund finance, subscription lines are well insulated because repayment rests on uncalled commitments from institutional investors. The vulnerable products are NAV facilities, private equity CFOs and rated note feeders, where fund valuations are the source of repayment. Fitch expects rating pressure to be limited to NAV facilities with tight headroom and to mezzanine or subordinated tranches with concentrated exposure to outsourcing, software, traditional media or legacy financial services.
For asset managers, Fitch does not expect AI to displace investment management. It expects dispersion. Firms that use AI well in research, portfolio construction and risk management could gain ground, while laggards risk client attrition. Fitch scores Investment Management Organizations at 40 but traditional and alternative Investment Managers at 20, reflecting limited asset-quality and business-model risk.
Structured Finance: Sructure Absorbs Shocks
GPU/AI compute ABS has the highest structured finance score at 50. Rapid chip turnover shortens the economic life of collateral, and repayment depends on the lessee. Deals backed by hyperscaler lease payments rest on committed cash flow from an investment-grade counterparty, while monoline compute providers are more vulnerable. Its rated GPU exposures carry no residual value exposure because the notes amortize within the contracted lease term. Elsewhere, overcollateralization, amortization and diversification can absorb a good deal of stress before ratings move. Fitch also flags that broad job displacement could raise delinquencies in consumer lending, credit card and student loan pools.
What To Watch Next
- Phase 2. Fitch will apply the test to individual issuers and transactions later this year. It does not plan to extend it to structured finance issuers.
- The 2028 software maturity wall. Refinancing, not near-term defaults, is the next test for BDC software portfolios.
- Scope limits. The test excludes second-order effects such as macroeconomic impacts, government tax revenue and AI-enabled cybercrime.
Concluding Thoughts
AI is not a uniform tide for credit, and it is not a threat to regulated banking as a whole. It widens the gap between resilient sectors and exposed business models. For investors and risk managers, the most important question is not whether AI will hurt credit, but who is lending to the sectors it hurts most and how much cushion will lenders hold. Phase 2 should start to answer that.
Leave a comment