Private credit dislocation

Thesis

Fourteen years of quantitative easing created ~$25 trillion in new central bank money that compressed yields on safe assets and forced insurance companies, pension funds, and sovereign wealth funds down the risk curve. Private credit was the destination: higher yields, mark-to-model stability, apparent decorrelation. The capital flowed into long-duration physical infrastructure financed against forward demand assumptions never tested through a downturn.

This is not a data centre story. Data centres are the largest and most visible pool, but the same financing pattern is being replicated simultaneously across fibre networks, tower companies, renewable energy, grid infrastructure, EV charging, and commercial real estate. Each vector shares the same characteristics: physical long-duration assets, private credit and ABS financing, forward demand assumptions, and BBB-adjacent credit ratings. The same global allocator base is on both sides of every deal in every geography. Geography does not create diversification. It creates correlation.

BBB OAS at 108 bps (28 Feb 2026) is pricing for a world where $7 trillion in committed infrastructure spending proceeds without friction. Semi-liquid private credit vehicles have already demonstrated structural failure under redemption pressure (Blue Owl OBDC II, Feb 2026). The BBB OAS is the direct measurement of the credit layer where the QE unwind concentrates.

The QE inheritance

Central banks created $25 trillion between 2008 and 2022. Private credit AUM grew from ~$500bn in 2015 to over $2 trillion by 2025. European private credit fundraising hit a record $65bn in the first nine months of 2025, up 14% year-on-year. As Basel IV pushes European banks out of long-duration lending, a structural shift from bank balance sheets (70% of European lending) toward private debt funds is accelerating.

Underwriting degradation followed mechanically. Too much capital chased too few deals. Direct lending spreads compressed from ~600+ bps to ~450 bps. Covenant-lite went from exception to norm. The capital funded projects that would not have cleared underwriting in a normal rate environment: long-duration physical assets with forward demand models complex enough that most allocators cannot independently verify them. They trust the GP model. The GP is incentivised to deploy.

The deferral mechanisms are QE-vintage: PIK toggles, amend-and-extend, NAV loans against fund portfolios, extend-and-pretend refinancing. Each postpones price discovery. Blue Owl OBDC II is what happens when the deferral runs out.

QE did not end. It was renamed. The Fed halted quantitative tightening on 1 December 2025 having reversed only half of the pandemic balance sheet expansion. In the same month it rebranded the Standing Repo Facility as "standing overnight repo operations," removed aggregate caps entirely, and moved to unlimited full-allotment format. The escalation since is stark: $29.4bn on 31 October 2025 (then the largest repo operation since the early 2000s), $13.5bn on 1 December (the second-largest liquidity injection since COVID), $74.6bn on 31 December (an all-time record), and $18.5bn on 17 February 2026, the day before Blue Owl froze OBDC II. By late December 2025, cumulative repo liquidity injections for the year had crossed $120 billion, far exceeding prior years. Bank reserves had fallen to ~$3 trillion by September 2025, the lowest since 2019. The system is not post-QE. It is in managed life support where the central bank intervenes at every routine stress point to prevent transmission. The Acharya, Chauhan, Rajan, and Steffen research programme (2024-2026) describes the mechanism: QE expands runnable liquidity claims (demandable deposits, repo liabilities), but QT fails to reverse them. The supply of reserves creates its own demand for reserves over time, ratcheting up the required size of the Fed's balance sheet. There is no clean exit. Every intervention sustains the deferral mechanisms (PIK toggles, amend-and-extend, NAV lending) by keeping system liquidity available. But each intervention also confirms the system cannot operate without it.

Current situation

Blue Owl as precedent

Blue Owl permanently halted quarterly redemptions at OBDC II on 18 Feb 2026, replacing them with mandatory return-of-capital distributions. The fund sold $1.4bn in loans at 99.7% of par. ~30% of NAV was distributed by end of March. OTIC (tech-focused BDC) saw redemption requests spike to ~15% of NAV. First clean demonstration that the semi-liquid wrapper fails under real pressure.

OAS levels (28 Feb 2026)

Index	OAS (bps)
ICE BofA BBB US Corporate	108
ICE BofA BB US HY	191
ICE BofA US HY Master II	312
ICE BofA Euro HY	274

Global IG spreads widened ~4 bps in the last week of February, the largest weekly move since early November. HY OAS at ~3% has been achieved only a handful of times: May 2007, briefly July 2021, intermittently since November 2024. Widening off a historically compressed base.

Contagion channels

Four transmission mechanisms connect a distressed infrastructure asset to broader credit markets.

Bank back-leverage. Banks fund private credit funds via leverage lines. Defaults breach covenants, creating bank-level exposure that was not underwritten as direct infrastructure risk.

Insurance and pension rebalancing. Heavy allocations to private credit mean solvency ratio degradation forces selling of liquid public equities. Stress transmits from private to public markets through portfolio rebalancing, not direct credit linkage.

Valuation opacity. Headline private credit default rate is sub-2%. Including selective defaults and liability management exercises, the true rate approaches 5%. PIK toggles and amend-and-extend mask credit deterioration until forced price discovery (redemption, refi, or regulatory intervention) exposes the gap.

PE-insurance pipeline. Private equity firms (Apollo, KKR, Blackstone, Brookfield, Carlyle, Ares) have built a three-cornered structure controlling a US life insurer selling annuities, an offshore reinsurer in Bermuda or the Cayman Islands, and the asset manager deploying capital into private credit, CLOs, and ABS. All three entities sit within the same holding company. The insurer sells annuities to US retirees, cedes liabilities offshore to its own captive reinsurer, and the reinsurer holds reserves under Bermuda GAAP rather than US statutory accounting, releasing capital that the asset manager redeploys into originating more private credit. The FSOC valued offshore reinsurance transfers at over $1 trillion as of 2025. This is not an adjacent risk. It is the funding source: annuity premiums from retirees directly finance the private credit origination across every infrastructure vector. Losses in infrastructure credit do not just hit fund NAVs and bank leverage lines. They hit insurer solvency ratios, triggering state regulatory action and forced asset liquidation that widens spreads further. PHL Variable entered rehabilitation in May 2024 with a $2.2 billion hole. Blackstone's $82 billion BCRED saw redemption requests surge to 7.9% of NAV in early March 2026, forcing the firm to inject $400 million of its own capital. The Bermuda Triangle structure makes opacity structural: offshore reinsurers file roughly 60 pages of disclosure versus thousands under US rules, and when the insurer and reinsurer are the same holding company, there is no genuine risk transfer.

These channels are identical across all infrastructure vectors and all geographies. The trigger can originate anywhere. The transmission mechanism is the same everywhere.

Infrastructure financing vectors

Seven parallel pools share the same financing structure, the same allocator base, and the same vulnerability to a narrative shift. Each is described separately because the risk characteristics differ. The contagion risk is that they are correlated.

Data centres

The largest and most concentrated pool. Global debt issuance nearly doubled to $182bn in 2025 from $92bn in 2024. The sector requires ~$870bn in new debt through 2030. Some borrowers are seeking loans exceeding 150% of construction costs. The Bank of England launched a review of data centre lending.

The risk sits below the hyperscalers. Google, Meta, Microsoft, and Amazon have the cashflow to absorb write-downs. Independent developers, colocation operators, and build-to-suit specialists have levered against forward lease commitments with thin equity cushions and construction execution risk. A hyperscaler that delays, downsizes, or renegotiates a lease leaves the developer with a drawn credit line, cost overruns, and no revenue. Construction loans on physical assets lack the PIK-toggle flexibility of software lending. ABS tranches rated against assumed lease cashflows face re-examination. The catalyst need not be large: one earnings call signalling a more measured build programme, or one credible research piece questioning AI monetisation timelines.

Fibre networks

Distinct risk profile from data centres. The core vulnerability is take-rate risk and overbuild, not construction-against-lease. European fibre absorbed EUR 8bn in infrastructure debt in Q4 2025 alone. Multiple operators are building parallel FTTH networks across France, UK, Germany, Spain, and Italy, each financing against subscriber projections that assume they win share from competitors building the same networks in the same streets.

The consolidation everyone expects has not materialised. Debt sits across fragmented operators with deteriorating unit economics. Fixed wireless and incumbent competition further compress take-rate assumptions. A take-rate miss on overlapping builds does not create a sudden crisis. It creates a slow grind of covenant breaches and amend-and-extend that ties up private credit capital and erodes returns, weakening the vehicles that also hold data centre and renewable exposure.

Renewable energy

European renewables attracted EUR 27bn across 143 deals in Q4 2025. Solar led at EUR 11.7bn. EUR 33bn in European solar financing closed across the full year. All financed via project finance, private credit, and green ABS against power purchase agreements and subsidy assumptions.

Three distinct risk layers. Political: subsidy regime shifts can retrospectively alter project economics (IRA rollback risk in the US, varying EU member state implementation). Curtailment: weak Nordic power prices in 2025 from strong hydro output and excess renewable capacity already hit investor returns, demonstrating that oversupply is not hypothetical. Grid interconnection: projects with drawn debt sit in multi-year queues, unable to generate revenue. After the EU Recovery and Resilience Facility expires at end of 2026, public co-funding drops sharply, pushing more financing to private credit.

Grid and power infrastructure

EUR 584bn per year in European grid investment needed (EIB estimate). Physical grid capacity for data centres is being financed against the same AI demand curve that underpins data centre economics. Circular dependency: if data centre demand slips, grid infrastructure is oversized; if the grid is not delivered on time, data centres cannot connect and revenue cases break.

Japan illustrates the constraint. $26bn+ committed by hyperscalers, but power connections in Tokyo take 5-10 years. Utilities investing JPY 150bn+ in substation upgrades on mismatched timelines. The April 2025 Iberian Peninsula blackout exposed grid fragility across southern Europe. Grid financing is the most directly coupled to the data centre thesis and the most exposed to circular failure.

EV charging infrastructure

ChargePoint, EVgo, and second-tier operators have levered against utilisation assumptions dependent on EV penetration rates that keep being revised downward. Infrastructure is in the ground but operating at sub-economic utilisation. Smaller in aggregate than data centres or renewables, but concentrates the same pattern: physical assets, forward demand assumptions, private credit filling the gap.

Tower companies

The three US tower REITs (American Tower, Crown Castle, SBA Communications) carry combined debt exceeding $65 billion at leverage ratios of 5-6x net debt to EBITDA. In Europe, Cellnex is the largest independent towerco with over 130,000 sites across ten countries and €20.8 billion in net debt at 6.28x leverage, rated BBB- by both S&P and Fitch. Vantage Towers (87,800 sites, ten European markets) was taken private by a Vodafone/GIP/KKR consortium, concentrating tower ownership within PE hands. Orange's TOTEM and Deutsche Telekom's DFMG (partnered with Brookfield) represent further PE and quasi-PE tower consolidation plays across France and Germany. In emerging markets, IHS Towers carries approximately $3.9 billion in debt with maturities concentrated in 2026-2027, exposed to Nigerian naira volatility and tenant concentration on MTN; MTN acquired the remaining 75% of IHS in February 2026 at a $6.2 billion valuation, re-integrating towers into operator balance sheets. Helios Towers is deleveraging toward 4.0x while Indus Towers (India, 200,000+ sites) is expanding into Africa. Tower ABS securitisation is a mature and growing market: SBA raised $2.07 billion in a single October 2024 closing; Vertical Bridge closed $1.94 billion in February 2026 with tranches rated from A down to single-B, the first single-B-rated tower ABS ever issued; Cellnex issued €1.5 billion in dual-series bonds in January 2026 to fund its 2026 maturities. The risk is distinct from data centres but shares the same financing architecture and the same credit layer. Tower revenue everywhere depends on long-term carrier leases with concentrated customer bases. In the US, DISH/EchoStar default on Crown Castle obligations exceeding $3.5 billion exposed this concentration risk directly. Crown Castle is executing a full strategic reversal, divesting $8.5 billion in fibre to deleverage, cutting 20% of its workforce, and targeting 6-6.5x leverage post-transaction. SoftBank's $4 billion acquisition of DigitalBridge (Vertical Bridge's parent) links the tower stack directly to the AI infrastructure narrative and its associated financing risks. The PE ownership pattern (GIP/KKR in Vantage, Brookfield in DFMG, SoftBank in DigitalBridge) mirrors the insurance contagion channel: infrastructure assets absorbed into PE holding structures with leverage optimisation as a core value-creation lever. Tower debt sits in the same IG and BBB-adjacent credit layer as the other vectors. A carrier retrenchment, 5G build deceleration, MNO consolidation reducing tenancy demand, or EM currency shock transmits through the same ABS and private credit channels globally.

Commercial real estate

$3 trillion+ in CRE debt matures by 2030. Private credit is stepping in as banks retreat. Property values down ~20% from early 2022. Office is a secular problem. CRE is the vector with the longest history of private credit involvement and the clearest existing stress. The mechanisms playing out now (extend-and-pretend, mark-to-model gaps, forced selling on refi) are the same mechanisms that will play out in data centres and renewables if demand assumptions disappoint. CRE is the leading indicator for the broader infrastructure credit cycle.

Global geography

United States

$200bn+ in data centre debt raised in 2025. Over $120bn moved off hyperscaler balance sheets via SPVs. $27bn in data centre ABS. Meta sealed a $30bn deal with Blue Owl for a single facility. Opportunistic and distressed debt funds have raised $100bn in two years, positioning for the turn.

Europe

Infrastructure debt hit ~EUR 80bn per quarter in Q4 2025. UK saw record GBP 68bn in 2025, with GBP 31bn from 13 deals. Europe's ABF market is EUR 4.2 trillion, but over 75% remains on bank balance sheets (non-banks at 13% versus 34% in the US). Basel IV is closing that gap. Data centre securitisation is nascent (EU has issued only EUR 0.8bn versus $63.6bn in the US since 2018) but scaling fast. Fibre overbuild risk, renewable financing, and the post-RRF public funding cliff are all most concentrated here.

China

Cloud providers expected to invest $70bn+ in 2026 (15-20% of US hyperscaler spend). RMB 500bn in government instruments targeting digital infrastructure, expected to lever up to RMB 7 trillion. Financing is more state-directed and bank-funded. Chinese banks have lent $200bn+ to the US over 25 years, including data centre financing. Broad infrastructure investment growth collapsed from 11.5% to 1.5% in 2025. If AI demand disappoints globally, Chinese banking absorbs the loss, but effects transmit via NPL ratios, EM lending appetite, commodity demand, and currency dynamics.

Japan

Third-largest data centre market globally. $26bn+ committed by hyperscalers. Power connections in Tokyo take 5-10 years. SoftBank concentrates AI infrastructure financing risk across multiple geographies (Stargate, sovereign cloud JVs with Oracle, already levered). A narrative shift hits Japan hard: the build is front-loaded against a grid that cannot deliver on schedule.

India and APAC

$150bn+ in AI and data centre capital announced or advanced across APAC in H2 2025. $15.2bn raised through 21 transactions by July 2025. India pursuing multi-gigawatt campuses at earliest stage with weakest enterprise demand base. Most execution risk: weaker grid, longer permitting, thinnest developer balance sheets. Earliest-stage projects break first.

Why BBB OAS is the right instrument

BBB is where QE-era reach-for-yield concentrated. Insurance IG mandates, pension LDI strategies, CLO warehouses: all in BBB because it was the highest yield available while remaining investment grade. Infrastructure project finance is structured to achieve IG ratings. Data centre ABS senior tranches are rated IG. Green bonds sit in IG allocations. When any vector comes under pressure, the repricing transmits into the BBB OAS composite.

The trigger does not need to distinguish between US data centres, European fibre, Japanese grid financing, or Indian greenfield builds. They all converge in the same credit layer. At 108 bps, there is no cushion. The distance from 108 to historically normal spreads (~150-170) is the repricing of the QE inheritance. From 170 to 200+ is the repricing of infrastructure demand assumptions. Beyond 200 is systemic.

Leading indicators

CCC OAS diverging from BB: credit differentiation returning signals cycle turn. CDS-cash basis widening on hyperscaler debt: smart money moving before public spreads. A third private credit gating event at a different manager: Blue Owl froze OBDC II, Blackstone injected own capital into BCRED; a third confirms systemic pattern. Hyperscaler earnings commentary on build pacing. Data centre ABS downgrades or construction loan impairments.

European infrastructure debt spread widening (Aviva tracks illiquidity premia quarterly; infrastructure spreads are moderately sticky, so movement is lagging but confirming). SoftBank credit stress or JV restructuring. Chinese infrastructure investment deceleration below 1.5%. APAC project delays or cancellations (India and Southeast Asia break first). CRE refi failures accelerating (leading indicator for the same mechanisms across other vectors). PE-affiliated insurer regulatory actions or additional captive reinsurance impairments (PHL Variable is the first; further state-level rehabilitation orders confirm pattern). Tower ABS spread widening or downgrade of lower-rated tranches (single-B tower ABS is new and untested through stress). Carrier capex guidance cuts or lease renegotiation announcements.

Dual trigger framework

Two independent paths lead to the same repricing event. Path A: liquidity withdrawal. The Fed either chooses or is forced to stop backstopping short-term funding markets. Inflation constraints, fiscal dominance, or political pressure remove the central bank's willingness to intervene. Deferral mechanisms lose their oxygen. PIK toggles, amend-and-extend, and NAV lending all require system liquidity to function. Remove it, and price discovery happens fast.

Path B: conviction shift. Allocators lose confidence in the asset class, the structures, or the forward demand assumptions underpinning infrastructure investment. A redemption wave overwhelms wrapper capacity regardless of available liquidity. Blue Owl and Blackstone BCRED are early signals on this path. Neither was a liquidity event. The money was there. The willingness to stay was not. The critical insight is that these paths compound. A conviction shift large enough to generate systemic redemptions forces the Fed to choose whether to backstop private credit directly, which it has not done and politically cannot easily do. These are not bank deposits or Treasury securities. And a Fed withdrawal could itself trigger the conviction shift by removing the implicit guarantee that the system will always be supported.

Today, confidence is being maintained and contagion is being prevented only through central bank action. The $18.5 billion repo operation the day before Blue Owl's announcement is not proof of system resilience. It is proof that the system required intervention at the exact moment a private credit wrapper failed. The question the thesis cannot answer is whether the trigger arrives via Path A, Path B, or their interaction. BBB OAS is agnostic to cause. It measures whether the credit layer is repricing, regardless of why. That is why it remains the right instrument.

Both trigger paths face the same obstacle: liquidity is being pumped through the system from multiple directions simultaneously. The monetary channel is the Fed's repo operations (unlimited capacity, record volumes, intervention at every routine stress point). The fiscal channel is the One Big Beautiful Bill Act (signed July 2025): $167 billion in cuts to corporate foreign profit taxation, permanent restoration of 100% bonus depreciation, retroactive R&D immediate expensing, permanent pass-through deduction, tips and overtime deductions, SALT cap increase to $40,000. Combined deficit impact runs to trillions over the budget window. Each dollar of fiscal stimulus and each repo intervention extends the runway for deferral mechanisms and delays the repricing this thesis anticipates. This is the honest limitation of the framework as a timing tool. The structural vulnerability is real. The seven vectors are correlated. The financing assumptions are heroic. But the system has access to monetary and fiscal channels that can sustain extend-and-pretend for longer than a structural analysis alone would suggest. The thesis does not predict when. It identifies the conditions under which a repricing becomes inevitable and specifies the instrument that will register it first.

Key judgement

The Blue Owl event is not BNP Paribas August 2007. It is the first demonstration that a known structural vulnerability fails under real pressure.

The conditions for broader contagion are global: compressed spreads priced for perfection, $7 trillion in committed infrastructure spending financed against heroic demand assumptions, the same allocator base on both sides of every geography, and valuation opacity masking credit deterioration across every vector.

This is the structural consequence of quantitative easing that has not ended. Too much capital was created. It went into long-duration physical assets financed against forward demand narratives never stress-tested. The underwriting was done by GPs incentivised to deploy, reviewed by allocators chasing yield targets, and rated by agencies modelling base cases. The issues were postponed via PIK toggles, amend-and-extend, and semi-liquid wrappers that promised liquidity they could not deliver. The central bank continues to provide the system liquidity that keeps these deferral mechanisms operational.

The trigger is not a black swan. It is a delayed payback on infrastructure, a narrative shift on AI monetisation, or a further gating event that establishes a pattern. It can originate in any geography or any vector, and it can arrive via either path: liquidity withdrawal or conviction shift. The evidence to date is that containment of each stress event has required active central bank intervention. That is not stability. It is managed instability. BBB OAS at 108 bps is the global thermometer. It provides no cushion.