Paid to Wait: The Case for a Private Infrastructure Credit Fund

TL;DR

On 29 January 2026 Norton Rose Fulbright published its annual cost-of-capital panel. MUFG's managing director for project finance in the Americas put a clean, fully contracted US construction loan at SOFR + 150 bp, unchanged on the year. On the same panel, Apollo's Apterra, a lender that syndicates loans in this market, priced borrowing-base loans to US renewables developers at SOFR + 450 to 600 bp.

Similar contracts, signed or close to signing, sit under both loans, and the developer's loan costs three to four times the spread. That gap is the area I'm interested in exploring.

In August I wrote that from 2028 the AI build-out becomes a financing question, not an engineering one. This is my answer to part of it.

My thesis: from 2026 to 2030, a USD 1bn private credit fund (the Fund) that lends against contracted infrastructure before it cash-flows, behind the banks that finance the build, can net about 6.6% a year in USD (modeled, unlevered, at SOFR of 3.87% on 1 October 2026). That's 9.4% gross. Net, it's about 270 bp over SOFR and about 110 bp more than senior infrastructure debt nets after fees.

The risk-adjusted case sits in the loss line. I charge the book 1.5% a year in credit losses, which assumes about 2.5% of it defaults every year. That's more than twice the worst-year default rate for senior project finance loans (about 1.1% in the first year after close, Moody's 1983 to 2018), and roughly seven to nine times the twenty-year average Moody's Analytics reports. The book can lose 2.7% a year, every year, and still match senior debt. The junior layer has no default history of its own, though, and on my construction-stress case the book nets 4.8%, below senior. The bet is on underwriting completion.

The return sits behind the build: developer borrowing-base and pre-construction loans, completion loans behind bank construction debt and, at the floor, sub-investment-grade contracted credit and bank risk-transfer positions. By geography: US power, storage and data centres first, then European power and bank risk transfer, then the defense and minerals supply chains that public lenders leave open. MENA is an optional sleeve beside a sovereign partner. Senior debt can't clear the bar after fees, and most mainstream holdco paper prices below the floor, so the Fund makes no senior loans and takes holdco paper only when it clears 450.

I'd change my mind if the January 2027 Norton Rose Fulbright panel puts the borrowing-base midpoint below SOFR + 450 bp (it was 525 in January 2026). That takes the base case to 6.1% and the edge over senior to about 70 bp. I'd also change my mind if MainePERS's table for 31 March 2027, the only LP-reported read on what this layer delivers net, shows Brookfield's third high-yield infrastructure debt fund below a 6.0% interim net IRR (an interim 7.3% a year earlier, on the GP's own cash flows). The vehicle has its own test: without anchor commitments of at least USD 300m by 30 June 2027, the right structure is a separately managed account.

How to read this

It's written for a prospective GP, its investment committee and anchor LPs, as a draft for discussion rather than for publication. "The Fund" is a hypothetical design sketched from public data. No GP is named, and nothing here comes from any fund's private documents.

---

I. Why now: the build-out is contracted, and banks are selling slots

There's a decade of contracted infrastructure to build, and banks will fund the senior debt on it cheaply and in volume. Whatever binds the cheap lenders (capital rules, ticket sizes or mandates), they all leave open the same slice: money that goes in before an asset produces cash, behind the senior lender.

I sized the demand two ways (my calculation). Published totals (IEA, Draghi, ReArm Europe, CHIPS, the Saudi Water Partnership Company's programme) give USD 190bn a year of new debt for contracted heavy infrastructure across the US, Europe and MENA. Countable projects (EIA's planned 2026 US additions, CBRE's construction pipeline, CHIPS awards, SWPC's project list) give USD 127bn, the number I use because it counts what lenders actually finance. Add refinancing and holdco lending, take out what banks, bonds and public lenders cover, and about USD 48bn a year is left for private credit.

++++

[ FIG. 01 ]

New-build private-credit residual19.7

Refinancing of operating contracted assets+10.0

Holdco and acquisition financing (estimate)+18.2

All private-credit-addressable debt47.9

Source: author's sizing model from EIA, CBRE, CRS, SWPC, IJGlobal, AEW and Preqin inputs. Worksheet in the companion data. Author's calculation.

Volume doesn't make the case on its own. The 30 largest infrastructure debt managers raised about USD 186bn over five years (Infrastructure Investor Debt 30), roughly USD 26bn a year for these three regions if, as I assume, about 70% of it targets them. At base the addressable pool is 1.84 times that capital. Count new-build alone and it's 0.76 times. Price settles it (Section II).

The demand is contracted, and the counterparties are states and large buyers. Britain's Allocation Round 7 contracted 8.4 GW of offshore wind in January 2026 on 20-year inflation-indexed CfDs. CBRE counts 7,481 MW of data centre capacity under construction in North America's primary markets, 80.4% of it preleased.

Banks aren't leaving. They lent just under USD 260bn of North American project finance across about 500 deals in 2025, 41% more than in 2024 (Refinitiv data cited on the Norton Rose Fulbright panel). Bank loans made up 80% of 2025's USD 1.05 trillion of infrastructure debt (Principal, on Infralogic data). Washington is loosening, not tightening: Fed staff put the March 2026 capital re-proposal at a 2.4% cut in CET1 requirements for the largest banks (Fed board memo).

A fund thesis built on "bank retreat" gets caught by the first LP who reads a league table.

What banks are doing is narrower, and more useful to the Fund. They're selling the slots their capital doesn't like. EU rules have put unrated pre-construction project finance at a 130% risk weight since January 2025, against 100% once a project operates (Simpson Thacher). The pre-2008 norm of project loans amortising over 18 years has given way to 5 to 7-year bank mini-perms that rely on refinancing for the tail (Proximo).

And banks now sell project finance risk outright. Two European project finance and infrastructure risk transfers referencing about EUR 4.8bn closed in October 2025 (A&O Shearman), and mBank and PGGM shared risk on a PLN 3.8bn Polish book that is about 75% wind and solar (mBank).

That last slot is a supply line. A buyer of junior and risk-transfer positions doesn't have to out-originate the banks.

Public lenders have started to work through private vehicles too. The Pentagon's Office of Strategic Capital launched a programme in July 2026 that routes critical-minerals loans through private funds (TechTimes, a press report). DFC put USD 600m into a critical minerals consortium beside USD 1.2bn of other money. The EIF anchored what the EIB calls Europe's first defense private credit fund with EUR 30m. Gulf sovereign funds more than quadrupled their private credit exposure to about USD 80bn between 2021 and 2025, mostly through incumbents' joint ventures and separate accounts.

The rate path suits a floating-rate lender. The FOMC raised the funds rate 25 bp to 3.75 to 4.00% on 16 September 2026, and coupons priced over SOFR rise with it. The cost shows up at the exit: with the 10-year Treasury near 5.17% in late September, the fixed-rate term-out that repays a completion loan at commercial operation gets harder.

---

II. The return sits behind the build, not in the contract

Read the ladder from the cheapest rung up.

++++

[ FIG. 02 ]

Loans, spread over SOFR (reported)Bonds and indices, spread over Treasuries

BBB US corporate bonds · OAS106

Senior private infrastructure debt index (Infra300)110

Bank construction loan · fully contracted US power150

Bank term loan · contracted, midpoint of 162.5 to 187.5175

Data centre construction loan · hyperscaler tenant, low-200s200

Mainstream holdco pre-construction loan · midpoint of about 350 to 450400

Borrowing-base loan · midpoint of 450 to 600525

Pre-construction loan at equity-replacement rates · midpoint of 600 to 800700

Sources: FRED BAMLC0A4CBBB (2026-10-01); SIPA infraMetrics debt indices factsheet (May 2026); Norton Rose Fulbright Cost of Capital 2026 Outlook (2026-01-29); JLL North America data center report (midyear 2026). Author's calculation of midpoints.

Senior contracted risk is priced like investment-grade corporate risk, because to a bank or an insurer it is. The BBB corporate option-adjusted spread was 106 bp on 1 October 2026. SIPA's model-priced index of senior private debt in 300 unlisted infrastructure companies sat at 110 bp in May 2026. Banks lend construction debt on fully contracted US power at SOFR + 150 bp and term debt at 162.5 to 187.5, with more than 100 active lenders competing for the paper.

Construction alone doesn't move the price. The step comes where the lender stands behind the build, or lends against assets that don't produce cash yet. Borrowing-base loans against developers' late-stage pipelines price at SOFR + 450 to 600 bp, and pre-construction loans went at what the panel called equity-replacement rates of 600 to 800. Most of that is a subordination and development premium. Part of it may also pay for contracts not yet signed and for the developer's own credit. The Fund lends only against signed contracts, so its borrowing-base paper may clear nearer 450 than 525, which is the 6.1% row in Section III.

The contract is the filter. It isn't the return.

Two neighbours price in the same band and give the Fund a second, better-documented source of return. Sub-investment-grade contracted credit carries the tenant's risk rather than completion risk, and its range is wide. The same panel reports data centre debt with CoreWeave as tenant sold at SOFR + 325 to 425. CoreWeave reports that its own compute facility for two non-investment-grade customers priced at SOFR + 450 bp in May 2026. JLL puts data centre construction loans for non-investment-grade tenants 200 to 300 bp wider than the low-200s that credit tenants pay. I book the sleeve at 400, inside that range, and give tenant credit its own floor of 400. Bank risk transfer has no public price at all; I assume 450.

Two layers don't clear the bar. Senior earns a fee, not a premium. Principal, an interested party, quotes 200 to 250 bp for investment-grade infrastructure issuance (Principal), and I use the 225 midpoint as a spread over SOFR. That's about 75 bp over the tightest bank pricing before fees, and a 60 to 75 bp fee consumes it. Mainstream holdco loans to developers price at "somewhere around 350 to 450 basis points over SOFR", in the words of Apterra, which has syndicated billions of them. At 1.0 to 1.46% losses they net about 5.2 to 6.6%. At the bottom they're below senior; at the top they're level with the Fund's base case, for a loan that ranks behind every project lender. Most new infrastructure credit funds will want to deploy there, because the paper is plentiful. It's also where the premium is being competed away, and it's why the Fund's floor sits at 450: holdco paper that clears it qualifies, and most doesn't.

One caveat travels with every number in the ladder. The borrowing-base and pre-construction ranges come from one practitioner panel, centred on US renewables developers, and the speaker's firm syndicates loans in this market, so read them as a seller's quotes. The tenant-credit prices sit in a neighbouring band, and none of them is a tracker. It's still the best dated public price feed this market has.

---

III. What a USD 1bn book earns, and why the risk-adjusted case holds

The base book holds 40% in developer borrowing-base loans at the panel midpoint of SOFR + 525 and 25% in single-asset completion loans at the same price. The rest is 5% in pre-construction loans at 650 (below the range's 700 midpoint) and 15% each in sub-investment-grade contracted credit at 400 and bank risk transfer at 450. The arithmetic is mine: SOFR of 3.87%, 0.50% a year of amortised upfront fees, 1.25% of management fee and expenses, and 10% carry over a 6% hurdle with no GP catch-up. For scale, the 2024 panel put total upfront fees on a project facility at 200 to 300 bp. Fees are design assumptions and don't represent any manager's terms.

Losses are charged sleeve by sleeve, and deliberately harsh. A borrowing-base loan is credit to a developer secured on late-stage assets, so I price it like the weakest Ba-rated corporate borrower: 2.43% default a year, the annualised twenty-year default rate of Moody's Analytics' weakest Ba bucket, at 60% loss given default, or 1.46% a year (my assumption). Single-asset completion loans carry 2.0% and pre-construction loans 2.5%, between that and the construction stress in Figure 3. The tenant-credit and risk-transfer sleeves carry 1.0%. Borrowing-base loans sit under their own 45% limit rather than the 40% construction cap. Each one is spread across a developer's pipeline, with the developer as obligor; a single-asset completion loan rides on one project reaching commercial operation (COD). About half the book is priced off an assumption rather than a quote: completion at the borrowing-base midpoint, risk transfer at 450 and pre-construction at 650. The downside rows below bound what that costs.

Senior nets 5.43% on my fee assumptions, and that's the bar. If the benchmark is 250 bp, the base case's edge is about 90 bp; at 200, about 140. SIPA's model-priced senior index sits at 110 bp, so a benchmark nearer it would only widen the edge; 225 is the conservative choice. In EUR (3M EURIBOR 2.617%) every row is about 1.2 points lower.

The catch-up row is why the term sheet has none: a full catch-up hands the GP half the edge.

My loss charge assumes about 2.5% of the book defaults every year, and the senior record says that's harsh. The junior record doesn't exist, which is what the stress case is for. Project finance loans default most in the first year after close, and Moody's 1983 to 2018 data put that worst-year marginal rate at about 1.1%. Over twenty years, Moody's Analytics reports infrastructure loan default rates of roughly 5.3 to 7.0% (Moody's Analytics), about 0.3% a year (my arithmetic). It calls them consistent with low investment-grade corporate issuers. Recoveries have been high: 84% on unrated infrastructure project loans over 1983 to 2021 (World Bank and GI Hub, from Moody's data). Junior lenders recover less, 64.6% against 76.3% for senior in the GEMs Consortium's emerging-market data (GEMs). That's why I use 60% loss given default rather than the roughly 35% those subordinated recoveries imply.

Here's the margin of safety. The book can lose 2.7% a year, every year, before it nets less than senior debt, and 4.3% before it nets less than SOFR (my calculation). After the loss charge it keeps about 350 bp of spread, against about 211 bp for senior private infrastructure debt after its own.

The risk-adjusted case rests on sub-investment-grade spreads on assets whose senior loans have defaulted at investment-grade rates. The Fund's job is to keep its own slice close to that senior record.

The risk sits in the tail. I ran the base book as 20 positions of USD 50m, five years, 200,000 trials, with default correlation from 0 to 0.24 (my calculation).

++++

[ FIG. 03 ]

Base bookConstruction stress on pre-cash-flow positions

Base book · independent defaults7.9

Base book · correlation 0.1214.1

Base book · correlation 0.2417.0

Construction stress · correlation 0.2449.2

Construction stress · independent defaults60.7

Model: author's simulation on the netting model's base book (worksheet in the companion data). Author's calculation.

The base book loses capital over five years in fewer than 2 paths in 1,000, even with correlated defaults. In the worst 5% of paths it still nets roughly 3.9 to 5.1% a year. The simulation tests the scatter around my loss assumption, not the assumption itself; the stress case does that. Put the construction-stress parameters on the pre-cash-flow positions and it trails senior about half the time, and in the worst 5% of those paths it nets under 1% a year, with about 3 in 100 losing capital. The return depends on underwriting completion, which is why the Fund caps construction exposure at 40% and lends only after sponsor equity.

Measured performance is thin, and an IC should hear it from the sponsor first. The only LP-reported number for a flagship high-yield infrastructure debt fund is MainePERS's 7.3% interim net IRR on Brookfield Infrastructure Debt Fund III at the end of March 2026, early in the fund's life and on cash-flow figures the GP provides. The same table marks one BID III co-investment at about 0.69 times contributed. I Squared's first infrastructure credit fund reported a 9.1% unlevered net IRR and no losses at the end of June 2024 (PSERS board materials), the manager's own figures for a young fund.

The wider evidence is less kind. SIPA's sub-investment-grade infrastructure debt index returned 3.43% a year over ten years to June 2025 (gross, model-priced) against 5.47% for US high yield. And private debt funds in general earn no significant abnormal return net of fees once risk is measured against both equity and debt benchmarks (Erel, Flanagan and Weisbach, NBER; mostly direct lending). Contract-backed collateral and 64.6 to 84% recoveries may change that result for this layer, but the edge over a direct-lending book is unproven, and I haven't priced it.

So who captures the premium? Senior contracted value accrues to sponsors and offtakers, through bank competition and concessional public lenders. The completion and junior premium is captured today by mega-managers writing large tickets and by specialists with captive anchors. Whether a new USD 1bn manager captures it net of fees and losses is not settled by public data.

---

IV. Where the opportunities are, by sector and region

The opportunity is widest where a creditworthy contract meets a cheap senior lender that won't go junior or early, on a ticket too small for the mega-funds: US power and data centres first. It's narrowest in the Gulf.

The base book maps onto this as roughly 70 to 75% US, 25 to 30% Europe and nothing in the Gulf until paper there clears the floor.

United States. The largest demand and the most volatile counterparty, the federal government. DOE is restructuring, revising or eliminating more than USD 83bn of Biden-era loans and commitments. The One Big Beautiful Bill Act ends wind and solar credits for projects placed in service after 2027 unless construction began before 5 July 2026 (IRS Notice 2025-42). So the Fund underwrites US projects on executed permits and contracts, never on grants, preliminary awards or conditional commitments.

Data centres stay eligible under a cap, because "hyperscaler lease" can hide both credit and tenor risk. S&P cut Oracle to BBB- on 9 July 2026 (heise). Meta's leases on its Hyperion campus run four years with extension options, backed by a capped residual value guarantee for the first 16 years, under bonds that amortise to 2049 (Meta).

Europe. Bank capital binds hardest here, and that's the supply line: completion loans banks would rather not hold through construction, and risk transfers on the books they've already written. The EIB sets the senior price, with a record EUR 100bn of group financing in 2025 (EIB). The structural risk is legal. The Commission ruled in March 2025 that paying Antin's EUR 112m award against Spain would be illegal State aid (TDM). Intra-EU investor-state arbitration is gone, so change-in-law protection on EU support schemes rests on the national courts of the state that changed the law.

Defense and minerals. These are the clearest structural gaps, because the cheapest lenders exclude them by rule. The EIB still excludes weapons and ammunition (EIB disclosure), and the EU's SAFE programme lends up to EUR 150bn to member states rather than to suppliers (Council of the EU). The Office of Strategic Capital's credit programme lists projects repaid mainly from federal grants or contracts as ineligible, at least under its inaugural rules. Its own deals now come with private capital alongside: on 31 July 2026 it made a conditional commitment of up to USD 820m to Performance Drone Works for drone component manufacturing (DoD). Frameworks are ceilings, though, so in this sleeve only firm, funded orders or capacity payments count as contracted.

Minerals fit where a government floor sits under the revenue, as with DoD's ten-year USD 110/kg NdPr floor for MP Materials. The floor is an appropriations risk: DoD's FY2026 request for the Defense Production Act Fund (USD 266m) sat below estimated floor payments (FAS).

++++

[ FIG. 04 ]

An assembly line on a floor plate with a press, a robot arm and an inspection gantry along a conveyor. A marker on the floor stripe moves through financial close, construction (the line's parts are installed), commissioning (the belt runs empty) and operation, when units roll off onto a pallet. No unit, and so no cash, leaves the line before the last stop.

An assembly line moves from financial close through construction and commissioning to operation, and no unit, and so no cash, leaves it before the last stop. Manufacturing counts as contracted only on firm, funded orders or a government price floor.

Source: the Fund's eligibility gate and term sheet (Sections V and VI). An illustrative plant, not a real project. Author's construction.

MENA. The senior market is cheap and long. Banks financed five Saudi solar projects with SAR 16.9bn for about 27 years, and Masdar's 2025 green bond came at 80 to 90 bp over Treasuries. Thirteen banks lent Masdar USD 5.1bn in July 2026 for its solar-and-storage project with state-owned EWEC, and Masdar put in the equity itself (Masdar).

Since 28 February 2026, Iranian strikes have hit GCC energy, water and industrial assets. By 19 March more than 25 companies operating in the GCC had applied force majeure (IISS). In March I wrote that the strikes had taken 19% of global LNG supply offline. So the Gulf stays an optional sleeve, sized to a sovereign partner. It takes junior tranches that clear the floor, on USD-pegged contracts only, and sticks to operating assets until twelve strike-free months have passed. Aqaba-Amman shows where the gap is real: IFC says long-term debt for the roughly USD 6bn project isn't sufficiently available from commercial lenders in Jordan.

---

V. Underwriting: what counts as a contract

The obvious way to define eligibility is a list of contract types. The 19 dated failures I could find from 2013 to 2026 say the list is the wrong unit. What protected lenders was who could stop paying, and what the termination clause left for the lender when they did.

The 19 cases sort into five kinds, and only one is an offtaker trying to stop paying; Oracle's notice, below, is the live example. Sponsors walked away before COD holding contracts: Ørsted ceased Ocean Wind 1 and 2 in October 2023 with DKK 28.4bn of impairments. Governments reversed permits: Interior suspended the leases of all five permitted offshore wind projects under construction in December 2025, and courts had enjoined every suspension by early February 2026 (Georgetown Climate Center). Suppliers failed under take-or-pay: Northvolt filed with USD 5.84bn of debt (Euronews) after BMW cancelled a roughly USD 2bn order (TechCrunch). Tribunals split on near-identical LNG contracts (Venture Global 8-K).

And a hyperscale tenant migrated toward junk. By mid-September about USD 18bn of bank loans tied to Oracle's Project Jupiter campus were quoted at 89 to 91 cents as syndication stalled (Reuters). The following week Oracle's force majeure notice to the developer became public, and its 2055 bonds traded at 77 cents (Axios). Oracle says the project remains on schedule (Insurance Journal).

Two protections worked. DOE's USD 475m loan to Li-Cycle was never advanced because a condition precedent required USD 262.7m of additional investment first; the company entered creditor protection in May 2025 (Insolvency Insider). And at the portfolio level, the type of revenue does separate outcomes.

++++

[ FIG. 05 ]

0.7%

Revenue-resilient · global

5.4%

Non-resilient · global

5.6%

Revenue-resilient · low and middle income

10.3%

Non-resilient · low and middle income

Source: World Bank and GI Hub, Infrastructure Monitor 2024 (2025-04-23), from Moody's data.

Most contracted project debt, PPA- and CfD-backed projects and leases included, sits in the higher bucket. The gate's job is to pick the best of it. It scores each contract on seven dimensions, 0 to 3 each: counterparty credit, payment obligation, tenor coverage, termination and change in law, currency, enforceability and completion support. A loan needs 14 of 21 with no zero. Junior and completion loans also need completion support (equity first or a sponsor guarantee), DSCR covenants and a termination payment that reaches the junior tranche.

That last condition is Carillion's lesson. Under the Royal Liverpool PFI termination formula the cost to complete is deducted from what lenders receive, so when it rose to GBP 293m the compensation fell to zero (National Audit Office, January 2020). Availability contracts default rarely and recover badly, 55% against 88% for other PPPs (Moody's 1983 to 2020, via the Infrastructure Monitor): low frequency, not low severity.

---

VI. The Fund: why USD 1bn, and what it has to prove

USD 1bn is the right size for this layer. With a 10% single-borrower limit it writes tickets of USD 25 to 100m, about 20 of them, below the USD 100m where Ares starts its infrastructure debt deals. Blackstone said in July that scale and large checks are prerequisites in private investment-grade credit (Q2 2026 call). A USD 1bn fund is big enough to diversify and small enough for one or two anchors to launch.

The term sheet is a straw man, and none of it comes from any fund's private terms. Fee levels lean on Dakota's 2026 public pension fee report, where private credit schedules run from 1.00% on committed to 1.5% on invested capital, with 20% carry over 8%.

The fees are below market on purpose. A new manager buys its anchor with terms. Charging on invested rather than committed capital, with the spread floor written into the investment policy, is my answer to the risk that fees reward deployment over spread. It's a plan, not evidence.

The competition is real and well funded. Brookfield closed its third high-yield infrastructure debt fund at over USD 6bn and had sold USD 4.66bn of the fourth, including parallel vehicles, by August 2026 (Form D). Blackstone's BGREEN III closed at USD 7.1bn.

Smaller entrants show how hard the raise is. I Squared's first credit fund closed at USD 715m against a USD 1bn target before scaling (ION Analytics). Rivage's junior European fund held a EUR 260m first close against a EUR 1.5bn target. Eagle Point's infrastructure credit vehicle had sold USD 100.3m against a reported USD 1 to 1.5bn ambition (Form D). None hit its target first time, and Rivage and Eagle Point sat below a fifth of theirs. A new manager starts from that base rate.

The gaps none of them covers well are mid-size completion and junior tickets on contracted assets, bank risk-transfer positions on project finance books, and the defense supplier tier. None of that is an edge until the GP can show a pipeline in it.

I tried running the eligibility gate over the 48 dated financings I collected, and public data can't do it. Seven put money into a subordinated, holdco or sub-investment-grade slot. Two disclose a price, CoreWeave's DDTL 3.0 at SOFR + 400 and DDTL 5.0 at 450, both on compute contracts. None discloses termination terms that reach a junior lender, and in one, Stegra, the junior debt has already been converted to equity. Capacity at the floor is the binding unknown. The first thing an anchor should ask to see is a dated pipeline of term sheets that pass the gate.

A winner needs four things, all testable: the ticket band, construction-monitoring and workout capability in house before the first completion loan, co-lending ties with public senior lenders, and an anchor in place of a track record. The anchor is the hard one. In 2025, managers incorporated before 2008 took 76% of infrastructure capital and those incorporated since 2020 took 5% (With Intelligence). Separate accounts carried USD 2.1bn of EIG's USD 4.0bn raise (Business Wire).

Sovereign funds say they want this paper: 69% of 73 surveyed plan to raise private credit and 51% are most interested in infrastructure debt (Invesco, 2024). But they want SMAs, fee breaks, co-investment and sometimes GP stakes. So the go or no-go line sits in the term sheet.

The other LPs follow in order: public pensions at a second close, after the first deals; insurers through a rated note feeder, with Solvency II's capital relief only on EEA and OECD assets (EIOPA); wealth channels last, because a semi-liquid wrapper on construction loans promises liquidity the assets don't have.

---

VII. The case against all of this

Now the case against, at full strength.

A new manager should not launch this fund. The junior premium is real on paper but compressing, and the managers best placed to harvest it already raise USD 4 to 7bn pools for exactly this risk. The 2025 anchor tickets in infrastructure debt went to established managers' new vintages: APG to Schroders, Nest to IFM. Banks sell their risk to the same mega-managers: Blackstone, Stonepeak and Clifford Capital bought the risk on SMBC's USD 3.2bn Asia-Pacific project loan book in late 2025 (IFR). Ares, itself a buyer, calls the risk-transfer market crowded and priced to perfection. Public lenders price strategic-industrial senior debt at Treasury and pick their private partners. The loss evidence is backward-looking senior bank-loan data, with no history at all for data centres, GPU collateral, defense suppliers or minerals processing. Nearly every recent failure sits in the Fund's target sectors. On my own stress case the book nets 4.8%, below senior.

That's the bear case, and most of it is true.

It kills any fund that holds senior or mainstream holdco paper, and it sinks a Gulf-core book. It doesn't rule out a completion and junior specialist with an anchor. Mega-fund ticket floors and public-lender exclusions leave a mid-size layer that still prices at SOFR + 450 to 800. The revenue model separates default rates too, 0.7% against 5.4% over five years, even if most contracted project debt sits in the higher bucket. The stress case is the honest risk, and it's an underwriting risk; equity first, conditions precedent and the 40% construction cap exist because of it.

If the GP can't show origination access and an anchor before first close, the bear case wins, and the answer is an SMA or nothing.

One historical analogue, and where it breaks. Project finance loans originated in 2006 to 2008 defaulted more often than other cohorts in Moody's 1983 to 2018 data, which reports 9.0 to 11.5% of them defaulting under the Basel definition. They were written in a boom and had to refinance into the 2008 freeze. 2025 was also a high-volume year, and completion loans depend on a take-out at COD. Where it breaks: those were senior bank loans, and the Fund can require equity first on every completion loan it writes. The analogue is a sizing argument, not a veto.

---

VIII. What I'm watching

Fastest-checkable first:

1. AI tenant credit, continuously to 30 June 2027. One more force majeure, deferral or cancellation notice on a signed data centre lease by a tenant rated BBB- or better, or Oracle's 2055 bonds below 80 cents for a full calendar quarter (77 after the September notice), and the data centre cap drops from 25% to 10%. That moves the book toward power and defense, inside the same thesis.
2. Strikes on Gulf power and water, monthly. Any recorded strike on a GCC power or desalination plant keeps the Gulf construction sleeve closed for twelve months.
3. The January 2027 Norton Rose Fulbright panel, by 31 March 2027. A borrowing-base midpoint below SOFR + 450 bp (525 in January 2026) takes the base case to 6.1% and the edge over senior to about 70 bp, too thin for the tail in Figure 3. That breaks the thesis. Between 450 and 500 the book nets 6.1 to 6.4%, and the Fund slows deployment. It's the most important date on the list.
4. MainePERS's private markets table, as of 31 March 2027. Brookfield's BID III below a 6.0% interim net IRR (7.3% a year earlier), or a second BID III co-investment marked below 0.8 times, would say the junior layer is losing more than I charge.
5. Anchors, by 30 June 2027. Below USD 300m committed, or terms that include a GP stake or fee-free co-investment above 50% of fund size, and the Fund launches as an anchored SMA instead.
6. The take-out market, quarterly. A 10-year Treasury above 5.5% for a full quarter (about 5.17% in late September) means re-underwriting every completion loan that depends on refinancing at COD.
7. The base rate, at each FOMC meeting. Every 100 bp off SOFR takes about 0.9 points off the net. If SOFR falls below 3% (3.87% on 1 October 2026), the target becomes a spread of about 270 bp over SOFR instead of a level.

---

Coda

A bank will lend against a signed contract at SOFR + 150 bp. A developer borrowing against a pipeline of such contracts pays SOFR + 450 to 600. The contracts underneath are similar, signed or close to it. What differs is who stands behind the build, and who gets paid last.

That gap is the price of subordination and of lending against assets that don't produce cash yet. A USD 1bn fund that underwrites completion well can collect it while charging itself a default rate more than twice senior project finance's worst year. It's being competed down from above, and the January panel will say how fast.

---

A note on the numbers

The sizing in Section I, the book returns, loss calibration and tail simulation in Section III, the midpoints in Figure 2 and the eligibility framework are my own calculations or constructions from public data; the worksheets are in the companion data. Everything else is someone else's number, named and dated in the text. Fees, carry and the risk-transfer price are assumptions and don't represent any manager's terms. Data freeze: 3 October 2026. The panel quotes, MainePERS's rows and the Moody's figures are checked against captured source text, not live pages. Four figures still need that live check before an IC relies on them: the Ba-rated default rate behind my 2.43% assumption, Moody's 2006 to 2008 cohort range, CoreWeave's DDTL 5.0 price and the Debt 30 window. I haven't re-checked them since the 3 October freeze, so treat them as reported until I have. The competitor table, LP map and full risk register a diligence pack would carry sit in the companion data.

References

Primary data

• Norton Rose Fulbright, Cost of Capital: 2026 Outlook (2026-01-29): loan pricing and North American bank lending; Cost of Capital: 2024 Outlook (2024-02-19): upfront fees. 2026, 2024
• Federal Reserve, FRED BBB OAS and SOFR (2026-10-01); FOMC statement (2026-09-16); staff memo on the capital proposals (2026-03-19).
• EIOPA, Solvency II qualifying infrastructure: the EEA and OECD condition.
• Scientific Infra & Private Assets, infraMetrics debt indices factsheet (May 2026) and The performance of infrastructure debt (December 2025).
• World Bank Group and GI Hub, Infrastructure Monitor 2024 (2025-04-23); Moody's Analytics, Examining infrastructure as an asset class (October 2020); GEMs Consortium via the EIB, Default and recovery statistics, private lending 1994 to 2024 (2025-10-07): default and recovery data.
• Public programmes: US DOE, loan restructuring (2026-01-22); Office of Strategic Capital, credit programme; IRS, Notice 2025-42; US State Department, critical minerals ministerial (2026-02-04); UK DESNZ, CfD Allocation Round 7 results (2026-01-14); Council of the EU, SAFE adoption (2025-05-27); EIB Group 2025 results and defense credit cornerstone; IFC, Aqaba-Amman disclosure (2026-02-25); UK National Audit Office, Carillion PFI hospitals (January 2020).
• SEC filings: Oracle 10-Q (quarter to 2026-08-31); Venture Global 8-K (2025-10-09); EIB exhibit (2025-03-21); Form D for Brookfield BID IV (2026-08-19) and Eagle Point (2026-08-06).
• LP and manager disclosures: MainePERS, private markets IRR table (2026-03-31); PSERS, I Squared GIC II board materials (2025-03-21); CoreWeave DDTL 5.0 (2026-05-18); Meta on Hyperion (2025-10-21); Ørsted on Ocean Wind (2023-10-31); mBank and PGGM (2025-10-22); Brookfield on BID III (2023-11-02); Blackstone BGREEN III (2023-08-10); Rivage REDI HR3 (2024-12-18); Ares infrastructure debt (accessed 2026-10-03).

Analysis and market

• Erel, Flanagan and Weisbach, Risk-Adjusting the Returns to Private Debt Funds, NBER w32278 (March 2024).

• Infrastructure Investor, Debt 30 (2026-07-01); With Intelligence, Infrastructure fundraising report 2025; ION Analytics on I Squared's credit fund (2026-03-09); Dakota, 2026 public pension fee report (2026-09-28); Invesco, sovereign investors and infrastructure debt (2024-12-02).

• Principal Asset Management, 2026 private infrastructure outlook and via InsuranceAUM (2026-02); JLL, North America data centers, midyear 2026 (2026-08-11); CBRE, North America Data Center Trends H1 2026.

• Simpson Thacher, Basel III endgame re-proposal (2026-03-25); Proximo, private credit as infrastructure's liquidity pocket; IFR, SMBC risk transfers (2025-12-19); A&O Shearman, EUR 4.8bn project finance SRTs (2025-10-29); Ares, risk transfer market (2025-11-11).

• Argaam on Saudi solar financing (2025-11-27); IISS, mapping Iranian strikes on the GCC (March 2026); Semafor, Gulf sovereigns' private credit (2026-04-09); TechTimes on OSC's fund-finance programme (2026-07-16); Federation of American Scientists on the DoD and MP Materials partnership.

• Company and case sources: Masdar financial close (2026-07); MP Materials and DoD (2025-07-10); DoD on OSC's Performance Drone Works commitment (2026-07-31); EIG SIDF VI (2026-09-09); APG and Schroders (2025-09-18); Nest and IFM (2025-05-20); Blackstone Q2 2026 call (2026-07-23).

• Failure cases: TDM on the Antin State aid decision (2025-03-24); Georgetown Climate Center on offshore wind lease suspensions; Euronews on Northvolt (2024-11-25); TechCrunch on BMW and Northvolt (2025-03-12); Insolvency Insider on Li-Cycle.

Prices and current conditions

• CNBC, 10-year Treasury near 5.17% (2026-09-27); Reuters, Project Jupiter loans at 89 to 91 (2026-09-18); Axios, Oracle's force majeure notice and 2055 bonds at 77 (2026-09-25); Bloomberg via Insurance Journal (2026-09-25); heise, S&P cuts Oracle to BBB- (2026-07-10).

Prior work in this series

• The Conversion Layer, Second Breakfast, August 2026.
• The Era of "Or" Arrived Four Years Early, Second Breakfast, March 2026.

---

Draft for discussion among investment professionals. Not an offer to sell or a solicitation of an offer to buy any security. Views are my own and do not represent PL Capital or Protocol Labs. Nothing here is investment advice. Do your own diligence.