Private Credit: Liquidity, Valuation Credibility, and Underwriting Repricing

1. Executive Summary

2. Market Structure and Where the Stress Is Concentrated

Liquidity Mismatch by Vehicle Type

Most private credit still sits in closed-end funds whose capital lockups broadly match loan maturities, which materially reduces classic bank-like run risk. The vulnerability has been reintroduced through retailization. BIS estimates BDCs now manage over $300 billion and represent 20% of the U.S. private credit market, while managers increasingly use evergreen and interval structures that allow periodic redemptions. FSOC reported that semi-liquid perpetual BDCs more than tripled in size from the end of 2021 to 2024-Q2, and the IMF warned in October 2025 that retail investors had become major contributors of new private-credit inflows. Those vehicles often carry 10%-40% of assets in marketable instruments, mostly leveraged loans, versus 1%-3% at publicly traded BDCs that do not permit redemptions. The structural problem is not simply that the assets are illiquid. It is that a liquidity sleeve can be exhausted or sold first, leaving the remaining investors with a progressively less liquid and potentially weaker pool of assets.

3. The Most Visible Cracks and Client Withdrawal Gates

The recent crack sequence has been unusually clear. BlackRock’s $26 billion HLEND received $1.2 billion of withdrawal requests in Q1, or 9.3% of NAV, and paid out $620 million, hitting the 5% threshold at which restrictions typically engage; 19% of its portfolio is in software. Blackstone’s $82 billion BCRED faced $3.7 billion of withdrawals, equal to 7.9% of the fund, and met all requests only by raising its standard limit from 5% to 7% and injecting $400 million from Blackstone and employees. Morgan Stanley’s North Haven Private Income Fund received requests equal to almost 11% of shares outstanding and returned only 45.8% of the tendered amount. Reuters reported, citing Bloomberg and a letter to investors, that Cliffwater’s $33 billion Corporate Lending Fund saw redemption requests of about 14% and capped repurchases at 7%. Blue Owl sold $1.4 billion of assets across 3 funds, returned part of the proceeds to investors, and permanently removed the quarterly withdrawal option from its smaller vehicle. This is not evidence that each of these portfolios is insolvent. It is evidence that the sector’s wealth-channel liquidity promise has met the first meaningful real-world redemption test.

4. Structural Stress Beyond the Gates

Confidence-Deterioration Scorecard

The gate headlines are only 1 expression of strain. A 2nd expression is the growing need to use balance-sheet or sponsor support to defend liquidity optics, as seen in BCRED’s sponsor capital and Blue Owl’s asset sale. A 3rd expression is public-market skepticism: Reuters reported that listed BDCs were trading at an average of 78 cents on the dollar of reported assets on March 12, down from 85 cents at the start of 2026 and roughly par in early 2025. A 4th expression is bank-side caution: JPMorgan has marked down some collateral values on loans to private credit groups, especially where software exposure is relevant, reducing leverage available to borrowers, although no material margin calls had yet occurred. A 5th expression is rising concern about underwriting and even fraud risk after the collapses of First Brands, Tricolor, and Market Financial Solutions, which Deutsche Bank explicitly referenced in its March 12 annual-report discussion of private-credit risk. Reuters also reported that a Rubric Capital letter alleged some BDCs may have temporarily shifted borrowings off balance sheet around reporting dates to appear less levered, although Reuters could not independently verify the practice or its scale. Taken together, these are not random idiosyncrasies. They are a broad deterioration in confidence around price discovery, collateral quality, leverage disclosure, and underwriting discipline.

5. Credit Quality and Cash Flow Stress

Credit Quality Deterioration Dashboard

On the underlying credit side, the official evidence has deteriorated, although not yet catastrophically. The IMF concluded in 2024 that more than 1/3 of private-credit borrowers had interest expense exceeding current earnings after the rate shock and that competition had already weakened underwriting standards and covenants. FSOC stated that default rates for the smallest borrowers, those with less than $30 million of EBITDA, had moved into double digits, and that nonaccrual rates had risen modestly even if they remained below recessionary levels. Fitch then reported that the U.S. private-credit default rate rose to a record 9.2% in 2025 from 8.1% in 2024, with 38 defaults across 28 borrowers in a 302-company monitor, concentrated in smaller issuers with $25 million or less in earnings. The floating-rate nature of the market matters here. Fitch noted that most of the loans were floating rate with minimal hedges, leaving cash flows highly exposed to still-elevated base rates.

An additional concern is that private credit has several internal shock absorbers that can postpone, rather than eliminate, loss recognition. FSOC reported that payment-in-kind income as a share of total BDC income has increased significantly since 2019 and warned that PIK elections can temporarily preserve liquidity while masking underlying credit problems and delaying loss recognition. That matters because a PIK-heavy book can look stable on reported yield and nonaccrual metrics even as the borrower’s true refinancing capacity deteriorates. The core issue is not that PIK is inherently improper. It is that an asset class marketed on steady income and low volatility can use PIK to preserve both optics longer than public markets would tolerate.

6. Software as the Pressure Point

Software Exposure Risk Matrix

Software has become the clearest pressure point because it sits at the intersection of private-equity ownership, enterprise-value underwriting, and AI disruption risk. Reuters reported, citing Morgan Stanley, that software accounts for about 16%, or $235 billion, of the $1.5 trillion U.S. loan market; 50% of software loans are rated B- or lower, more than 80% are issued by private companies, nearly 78% are sponsor-backed, roughly 30% mature by 2028, and 46% come due within 4 years. Reuters also cited BNP estimates that software and services exposure in private credit is around 20%. More importantly, the software equity selloff has affected credit math directly. Reuters reported that software stocks were down 22% since January, increasing loan-to-enterprise-value ratios on associated credits. Reuters also cited KBRA data showing that private-credit software borrowers carried average leverage of 7.4x EBITDA versus 5.9x across a broader loan pool, while sales growth had slowed to 10% from 18% a year earlier. This is the mechanism by which an equity-market narrative becomes a private-credit mark and refinancing problem.

The software problem should not be overstated into an immediate systemic default spiral. Morgan Stanley’s own view, as reported by Reuters, was that the risk of large, systemic disruption across software is limited in the near term and that a near-term spike in defaults is unlikely, even if loan price volatility persists. The more likely sequence is slower moving: wider spreads, tighter covenants, lower leverage multiples, tougher refinancing terms, fewer sponsor-backed software deals, and progressive mark pressure rather than an overnight wave of payment defaults. That distinction matters for both equity timing and credit-tranche selection.

7. Valuation, Mark-to-Model, and Why Reported NAVs Are Under Question

The valuation debate is the core analytical issue. Official bodies have already framed the problem clearly. FSOC stated that the absence of a substantial secondary market and limited transparency around valuation practices have raised concerns about stale valuations, that lenders determine fair value and nonaccrual status using unobservable inputs, and that the same or similar loans can be carried at different valuations and accrual statuses across BDCs and private funds. FSOC further warned that managers may be incentivized to maintain high valuations and delay recognition of losses. The IMF made the same point from a different angle, observing that private loans are often marked only quarterly using models and, despite lower credit quality, tend to show smaller markdowns in stress than traded leveraged loans. Those facts do not prove universal misconduct. They do show that the reporting process is structurally predisposed toward smoothing and delay. That smoothing also mechanically understates realized volatility and correlation, meaning historical Sharpe ratios and diversification statistics for private credit should be treated as partially accounting-driven rather than purely economic.

The academic evidence is more nuanced and points toward a balanced conclusion. The 2025 UNC paper on valuation discipline found that at least 76% of BDCs use third-party appraisal and 58% are required by their own creditors to do so. The paper also found that appraised loans show lower serial correlation in valuation updates and that deviations from appraisals are rare and small in normal times. However, the same paper found that deviations increased markedly during the COVID-19 period and that nearly 5% of loans were marked more than 5 points above the recommended range during that episode. It also found that lead lenders can receive higher appraisals even on the same loan because they possess more soft information from amendments, waivers, and renegotiations. The most defensible conclusion, therefore, is that private-credit marks are neither outright fiction nor reliable executable prices. They are model-based fair-value estimates that become meaningfully less robust precisely when investors care about them most. The SEC’s Division of Examinations has already highlighted valuation of illiquid assets as a live exam priority in 2025.

From an investment perspective, “not being marked properly” is best translated into 3 more precise statements. First, reported NAVs are almost certainly too smooth relative to public-market clearing prices. 2nd, dispersion across managers is large enough that manager marks are not interchangeable proxies for economic value. 3rd, sectors undergoing fast enterprise-value repricing, especially software, are where the gap between internal marks and market-implied values is most likely to be widest. That is why public BDC discounts and bank remarking matter. They are external challenge functions applied to internal models. The fact that Blue Owl sold a $1.4 billion loan portfolio at 99.7% of par and at the same level as its marks shows that some portfolios can validate near book. The fact that listed BDCs trade at 78 cents on the dollar of reported assets and JPMorgan has selectively hair-cut collateral shows that the market does not believe that validation is sector-wide.

8. Interconnectedness and Systemic Risk

The system-wide question is whether this remains a contained credibility event or evolves into a broader credit contraction. The Federal Reserve’s 2025 note is still the best anchor for the direct-bank-risk question. It found that committed lending by large U.S. banks to private-credit vehicles rose from about $8 billion in 2013-Q1 to about $95 billion in 2024-Q4, with utilized amounts at $56 billion. In a hypothetical full draw of the remaining undrawn commitments, the incremental draw would be about $36 billion, or roughly 2% of Y-14 banks’ CET1 capital, with only about a 2 bp hit to aggregate CET1 ratios and about a 1 percentage point hit to aggregate LCR. The Fed’s bottom line was that direct financial-stability concerns from this channel appear limited. That conclusion should be taken seriously.

The caveat is equally important. The same Fed note warned that liquidity demands from private-credit vehicles could be correlated with other NBFI drawdowns and exceed historical experience. The IMF estimated in April 2025 that the identified portion of global bank exposure to private-credit vehicles already exceeded $500 billion and likely surpassed 25% of total AUM in private-credit funds. FSOC estimated that committed bank facilities to BDCs had risen from $42 billion in 2020 to almost $117 billion by Q3 2024, while the IMF’s October 2025 GFSR warned that bank-private-credit partnerships had not been tested over time and could weaken underwriting and monitoring. The same IMF chapter also highlighted rising insurer exposure through structured instruments, feeder notes, collateralized fund obligations, and other private-fund links. The implication is that the likely contagion path is not a 2008-style bank solvency shock. It is a slower-moving tightening in liquidity, haircut practices, risk appetite, and correlated asset sales across a larger NBFI-financial complex.

9. Equity Market Implications

The 1st-order equity losers are listed BDCs and alternative managers whose earnings mix depends heavily on retailized, semi-liquid credit franchises. Those models face slower fundraising, higher redemption-management costs, lower confidence in reported NAVs, and lower market multiples applied to fee streams backed by illiquid assets. Reuters reported on February 6 that the software shock erased almost $60 billion of market value from alternative managers and that names such as Ares, Blackstone, Blue Owl, Carlyle, Apollo, TPG, and KKR fell 7%-14% during that week. If the stress remains contained, stronger managers with locked-up institutional capital may ultimately benefit from wider spreads and less competition, but near-term equity pricing is likely to remain dominated by redemption headlines and valuation skepticism.

Beyond the alternative-asset complex, tighter private credit should pressure the multiples of sponsor-heavy, leverage-dependent sectors. Direct lending has become the dominant funding channel for many middle-market private-equity-backed companies. If that channel is repriced, the consequences are lower leverage availability, wider spreads, tighter covenants, fewer dividend recaps, slower LBO activity, slower exit markets, and weaker transaction comps. The public-equity sectors most exposed are sponsor-heavy software, IT services, selected healthcare services, consumer finance, and parts of the industrial middle market. For banks, direct capital impairment still appears limited, but warehouse lenders and bank-partner models face earnings and risk pressure from tighter haircuts and lower transaction volumes. For insurers, the IMF’s warning on structured private-credit exposure is an underappreciated channel. If the stress stays contained, the broader index effect is more likely to resemble a style rotation toward balance-sheet strength than a generalized equity bear catalyst.

10. Generative AI Infrastructure, Technology, and Power

Weakness in private credit is unlikely to stop the core generative-AI infrastructure buildout. The reason is straightforward. The largest builders are not marginal borrowers. The BIS estimated in January 2026 that spending on data centres and IT manufacturing facilities had already reached 1% of U.S. GDP by mid-2025, with total IT investment at 5% of GDP and semiconductor-plus-data-centre capex contributing an average 0.4 percentage points to GDP growth over the prior 3 years. BIS also noted that free cash flows had recently lagged capex, which helps explain why even very strong issuers are moving toward debt funding. Reuters reported on March 11 that Amazon was seeking about $37 billion in bonds and drew about $126 billion of peak demand, that Oracle expected to raise $45-$50 billion in 2026, that Alphabet had completed a global $31.51 billion debt raise including a 100-year bond, and that Meta had previously filed to raise up to $30 billion for AI infrastructure. The core hyperscaler buildout is therefore far more dependent on internal cash flow, equity valuation, and public bond-market access than on marginal private-credit availability. Semiconductors, networking, and core cloud-enablement supply chains are consequently more levered to hyperscaler capex than to private-credit liquidity.

Private credit nevertheless matters at the margin, and the marginal project is where the buildout can slow. BIS estimated that outstanding private-credit loans to AI-related sectors had risen from near 0 to more than $200 billion and from less than 1% to almost 8% of total private-credit loan volumes, with potential to reach $300-$600 billion by 2030. BIS also noted that AI-related private-credit loans are substantially larger than loans in other sectors, averaging $169 million versus $90 million, while maturity and spread are similar at 4.7 versus 4.8 years and 6.2 versus 6.1 percentage points. Reuters reported in December 2025 that AI data-centre and project-financing deals had surged to $125 billion in 2025 from $15 billion in 2024, and that Morgan Stanley estimated private credit could provide more than half of the $1.5 trillion needed for the data-centre buildout through 2028. BIS further warned that some AI financing structures may mask leverage or create circular financing inside the AI ecosystem. That combination implies that private credit is increasingly important not for the biggest incumbents, but for the 2nd derivative of AI expansion: bridge loans, mezzanine debt, developer financing, GPU-leasing platforms, non-hyperscaler cloud providers, data-centre developers without investment-grade tenants, and securitized structures backed by data-centre rents. Those are the segments most exposed to a rise in required returns and a fall in available leverage.

The operational transmission mechanism is likely to be financial rationing rather than outright shutdown. In a weaker private-credit market, more projects will require pre-leasing, stronger counterparties, larger sponsor-equity checks, tougher cash-sweep provisions, lower advance rates on residual asset values, and wider spreads on construction or holdco debt. The projects most likely to lose financing are those whose underwriting depends on optimistic future utilization, residual GPU values, or aggressive tenant assumptions. Some financing can migrate into ABS or rent-backed securitizations, but that changes the wrapper rather than removing the underlying project risk. The best-positioned projects are those with hyperscaler counterparties, long-term contracts, high-quality power access, and clear take-out financing paths. This will not prevent AI infrastructure from being built. It will make the buildout narrower, more expensive, and more concentrated.

11. Power Sector Implications

In power, the dominant conclusion is that financing risk is secondary to physical bottlenecks, but it still matters at the margin. The IEA projects global data-centre electricity demand to reach around 945 TWh by 2030, just under 3% of global electricity consumption, and projects electricity generation serving data centres to rise from 460 TWh in 2024 to more than 1,000 TWh in 2030. In the United States, natural gas already supplies more than 40% of data-centre electricity and is expected to provide the largest incremental supply through 2030, adding more than 130 TWh, while renewables add about 110 TWh and technology companies have plans to finance more than 20 GW of SMRs. Reuters, citing the EIA, reported that U.S. power consumption hit a record 4,198 billion kWh in 2025 and is projected to rise to 4,256 billion kWh in 2026 and 4,364 billion kWh in 2027, with AI and data centres among the drivers. The secular demand signal is therefore not in doubt.

What private-credit weakness can do is change which power projects get financed and on what terms. Merchant or quasi-merchant generation, behind-the-meter gas, co-located microgrids, and speculative interconnection-led developments are more sensitive to a pullback in private credit than regulated utilities or large contracted infrastructure vehicles with public debt access. The IEA explicitly warns that long grid-connection queues mean that, in higher-demand scenarios, much of the incremental supply beyond the base case will be met by fossil fuels because projects that are dispatchable and financeable can move faster than ideal low-carbon buildouts. The likely equity implication is relative support for regulated utilities, transmission owners, grid equipment suppliers, large gas infrastructure, and contracted power developers, alongside more selective financing conditions for smaller private developers. In other words, weaker private credit is less likely to kill power demand than to increase concentration and raise the strategic value of existing balance-sheet capacity, existing interconnections, and established utility franchises.

12. Bottom Line

The current private-credit drawdown is best understood as a repricing of liquidity promises, underwriting discipline, and valuation credibility. The visible crack map now includes gated or prorated redemptions, sponsor-supported tenders, asset sales to defend liquidity optics, record default rates among smaller borrowers, rising PIK usage, widening public discounts to reported NAV, selective bank remarking of collateral, and renewed focus on underwriting and fraud risk. None of that yet proves a generalized collapse of the asset class. It does show that the most fragile structure is the combination of illiquid middle-market lending, periodic retail liquidity, model-based NAVs, and layered bank funding. For equities, the likely result is continued de-rating pressure on listed BDCs and wealth-channel alternative managers, with spillovers into sponsor-heavy cyclicals and software-linked credit exposures. For generative AI infrastructure, the likely result is not cancellation of the core buildout but a more concentrated capital stack: hyperscalers, investment-grade issuers, regulated utilities, and best-contracted developers continue to build, while marginal projects become slower, more expensive, and more equity-intensive.