Sponsorship: Private Debt News reaches institutional investors, credit professionals, and LP decision-makers. Early sponsor rates available. Contact us here or reply directly to this email.
Carlyle published a white paper this week with a number that should stop every allocation committee in the industry. Private credit may need to provide roughly $1 trillion to finance AI compute infrastructure. That’s more than half of everything the asset class currently manages.
The paper’s conclusion is blunter than the math. Failing to set clear concentration limits on AI compute “could prove to be the biggest mistake of all.” Co-president Mark Jenkins drew the comparison nobody in the industry wants to hear. Software accounted for roughly half of PE deals between 2020 and 2022 because recurring revenue looked stable and uncorrelated. Lenders piled in, everything appeared diversified, and then a shared technological threat hit every name at once. Jenkins sees the same structure forming in AI, where financings that look spread across data centers, chips, and SPVs ultimately depend on seven or eight counterparties. “In a crisis scenario, all of those things are going to really, really matter. They don’t matter when everything’s going well.” Some Carlyle investors are already asking for 8-10% caps.
The warning arrived the same week Apollo expanded its SoftBank loan to $9.2bn for the OpenAI bet, Fortress lent $265mm to an AI hardware maker, and PaleBlueDot AI went looking for $600mm from Brookfield to buy chips for Korea.
Apollo, meanwhile, started rolling out daily prices across its $850bn credit business. Asset-level marks for direct lending, ABF, and opportunistic vehicles arrive October 30. The timing wasn’t accidental. The SEC issued a “critical reminder” on private asset valuation on Monday, the kind of statement that creates no new rules and signals an enforcement priority. Apollo’s John Zito framed it as convergence with public markets. Rivals privately call it an unwelcome source of mark-to-market volatility. Both can be right.
The redemption picture mostly leveled off, with one exception. Blue Owl Technology Income Corp got requests for 39% of shares, up slightly from 38.1%, while its larger credit fund improved to 16.8%. BCRED held at 10%, Apollo fell to 14.7%, Ares to 13.1%. Goldman avoided its cap again. In Australia, Metrics Credit Partners suspended redemptions outright and S&P put four of its funds on watch negative.
Funding costs are rising with Treasury yields at two-decade highs. Bain Capital’s BDC couldn’t tighten its bond from initial talk at T+290, the first time since 2020 that two high-grade issuers failed to improve terms on the same day. Prospect Capital is discussing 8% on its first junk bond since 2021.
Key Market Themes
1. Carlyle Draws the Line on AI Concentration
Carlyle’s white paper estimates AI-related capex will top $5 trillion through 2030, with private credit potentially on the hook for about $1 trillion of compute financing. The funding takes several forms, including data center construction, power, chip-backed loans, and SPV lending. Unlike software, the paper argues, AI credit risk is more speculative, more correlated with the broader economy, and sitting inside structures that haven’t been tested.
Jenkins told Bloomberg TV that investors are requesting allocation limits of roughly 8-10%, particularly on the syndicated side. “We’re in a period where the revenue model to date is uncertain. In such an environment, it’s really hard for us as credit investors to say, ‘well, we’re all in.’” Carlyle isn’t exiting the space. “We want to take the risk, but we want to do it in a balanced manner.” The core problem he identified is that nobody can say yet where AI profits will accrue, whether to chipmakers, data centers, or application builders.
Diversification that isn’t
The software analogy is the right one and the industry should sit with it. Lenders in 2021 believed they held diversified portfolios of subscription businesses across healthcare IT, marketing tech, security, and ERP. They held one bet on software multiples and one bet against technological disruption. AI compute looks the same. A chip-backed loan in Korea, a Texas data center lease, and a hyperscaler SPV all resolve to the same question of whether a handful of frontier labs generate enough revenue to pay rent on hardware that depreciates faster than the lease. Jenkins’ point about seven or eight ultimate counterparties is the number that matters. A $1 trillion book with eight obligors isn’t an asset class. It’s a trade.
2. Apollo Goes Daily as the SEC Sends a Reminder
Apollo began publishing daily prices across its credit platform on Thursday, extending a program that started July 1 with investment-grade assets in its fixed income replacement strategy. Direct lending, ABF, multi-credit, and opportunistic vehicles follow, with asset-level pricing from October 30. Prices are benchmarked to public comparables, sector spreads, rate observations, and borrower financials, assessed the prior business day and shown on an investor portal. Apollo stressed they’re not clearing prices. The firm’s secondary desk has traded over $30bn since its 2024 launch.
The SEC’s Monday statement reiterated that the absence of market quotes doesn’t relieve managers of their obligation to estimate fair value. KKR’s Christopher Sheldon was dry about it. “Maybe it’s always helpful to have reminders. Often, people need to hear things multiple times.” Len Tannenbaum was more direct. “What impresses me about the SEC statement is its clear reminder that a lack of timely information does not relieve management of its responsibility to estimate fair value.”
The reference price problem
Apollo publishing daily marks on $850bn creates a benchmark that every other manager now has to explain their deviation from. When a loan sits in Apollo’s book and a competitor’s, and Apollo’s price moves on Tuesday while the competitor’s stays flat until quarter-end, the competitor has a conversation to have with its board and possibly with the SEC. That’s the real impact, and it’s why rivals are uneasy. The volatility complaint is honest but beside the point. Loparex went from 88 to 5 in nine months under quarterly marks. The volatility was always there. Apollo is just reporting it faster, and the SEC’s reminder reads like a notice that the quarterly-mark defense has a shelf life.
3. Blue Owl’s Tech Fund Stays at 39%
Blue Owl Technology Income Corp received requests to pull 39% of shares in the third quarter, up from 38.1% and only modestly below the 40.4% peak in the first quarter. The roughly $5bn fund capped at 5% again. Blue Owl’s letter attributed the stickiness to “the disconnect between the market’s fears of AI disintermediating software and OTIC’s resilient credit fundamentals” and said most requests came from investors rejoining the queue. OTIC will have returned about $446mm, or 35% of original requests, within six months.
The $35bn Blue Owl Credit Income Corp improved to 16.8% from 18.8% and 21.9% before that. Both funds have returned over 9% annualized since inception and expect higher base rates to help earnings. Across the industry, BCRED held at 10%, Apollo Debt Solutions fell to 14.7% from 16.8%, and Ares Strategic Income dropped to 13.1% from 14.4%.
The specialized mandate is the problem
Everyone else’s queue is shrinking. OTIC’s isn’t. The letter blames the market’s AI fears rather than the fund’s credit, and on non-accruals that’s defensible. But a fund built to concentrate in software lending can’t diversify its way out of a sector-wide reassessment, and investors have concluded the mandate itself is the risk. At 39% requests against 5% payouts, roughly ~1/3rd of the fund wants out and will keep resubmitting for the next six quarters. That’s a vehicle in slow liquidation regardless of what the loans do. The lesson for the rest of the industry is about product design. A single-sector evergreen fund sold to wealth clients was never going to survive the sector falling out of favor.
4. Australia Gates Outright
Metrics Credit Partners, one of Australia’s largest private credit managers, suspended redemptions on several funds, with Perpetual as responsible entity issuing the statements. S&P placed four Metrics funds on watch negative. The move follows Bathla Group’s August collapse owing A$3.4bn mostly to private lenders, and comes as Morningstar Australia reports private credit topped all asset classes with over A$5bn of net inflows last financial year while listed equities saw outflows.
Zenith’s Dugald Higgins described the dynamic. “It’s a negative feedback loop. People get nervous and sell and so therefore people get nervous and sell.” He added that another rate hike would push investors further toward cash regardless of how safe a fund looks. Morningstar’s Thomas Dutka flagged the overlooked liquidity lever, the bank credit facility, and noted HSBC had already told riskier fund clients globally that facilities wouldn’t renew.
The feedback loop has no 5% cap
The US industry’s quarterly 5% cap is a brake. Australia’s monthly-redemption, short-notice structures for similar loan books have no equivalent, which is why Metrics went from open to suspended rather than from 5% to 5%. Dutka’s point about credit facilities is the part US managers should read closely. Every BDC that expanded its revolver this summer is relying on a bank that renews annually and can decline. HSBC already declined for some. The gates in the US work because the facilities behind them are still open. That’s an assumption, not a structure.
5. BDCs Pay Up for Debt as Yields Hit Two-Decade Highs
Bain Capital Private Credit sold $350mm of notes due 2031 at T+290 with no tightening from initial price talk. The same day Campbell’s priced a hybrid at 8.5% with no improvement either, the first time since 2020 that two high-grade issuers failed to move pricing from talk on one day. A slump in Paramount Skydance’s jumbo bond had already rattled desks. Prospect Capital is discussing roughly 8% on its first junk bond since 2021.
In Europe, banks are preparing to undercut private credit on the €3bn-plus Cary windshield-repair buyout, with Jefferies circulating materials to both camps. Banks recently offered Sunday Natural at E+350 for a B2 credit.
The spread between funding and lending is closing
A BDC borrowing at T+290 and lending at S+475 was a good business when Treasuries yielded 4%. With Treasuries at 20-year highs and bond buyers refusing to tighten, the cost of fund leverage is rising faster than the coupons on the underlying loans, which are already being refinanced away at S+275 by the Mercers of the world. Bain’s deal priced. It just priced at the first number, which in this market counts as a warning. The Cary process shows the other side. When banks can offer a B2 at E+350, private credit either matches and compresses its own margin or loses the deal to a cheaper bid.
6. The Money Keeps Arriving, Just Not for Software
Audax Private Debt closed its third direct lending fund with $5.4bn in commitments and $10bn of total deployable capital including leverage and SMAs, roughly double its predecessor. More than 100 institutions participated over 21 months. CEO Kevin Magid said LPs repeatedly asked about software exposure, which the firm keeps under 10%. The fund has deployed $1.5bn into distribution, industrial manufacturing, business services, and pharma, and avoids restaurants, retail, energy, and travel.
Oaktree raised $2bn for its first dedicated ABF fund, targeting unrated opportunities that fall outside insurer constraints. “We see an opportunity to get overpaid for comparative credit risk, to get premium returns in that unrated segment of the market,” said head of strategy Jennifer Marques. BC Partners is seeking $2bn for its fourth special opportunities fund after the prior vintage hit a near-20% net IRR. Closed-end private debt funds raised roughly $133bn through July against $186bn for all of 2025.
Where the institutional dollar goes now
Audax doubling its fund at under 10% software while OTIC bleeds at 39% tells you what the institutional buyer wants. Plain middle-market lending to sponsor-backed industrials, no evergreen structure, no retail queue, and a sector mix that avoids the obvious disruption trade. Oaktree’s ABF raise is the other half. Insurers took the rated ABF market and compressed it. The unrated remainder, which Oaktree pegs at a few hundred billion, is where the premium survived. I’d read both closings as the same signal. Institutional capital hasn’t left private credit. It has left the products and sectors that retail bought at the top.
Deals of Note
SoftBank - Apollo expanded loan to $9.2bn from $5.4bn to fund the OpenAI investment
SourceCode - Fortress provided $265mm private credit financing for the AI hardware and HPC cluster maker
PaleBlueDot AI - In talks with Brookfield and others for $600mm to buy chips for a South Korea site
Cary - KKR, Blackstone, and Warburg considering bids for the €3bn-plus windshield repairer; banks preparing cut-rate financing to beat private credit
Akquinet healthcare - Lincoln running lender education for a carve-out of the ~€20mm EBITDA division from DBAG-backed parent
Audax Direct Lending Solutions Fund III - $5.4bn commitments, $10bn deployable, under 10% software
Oaktree ABF Fund I - $2bn for unrated asset-backed lending across equipment, transportation, consumer, and infrastructure
BC Partners SOF IV - Seeking ~$2bn, predecessor at near-20% net IRR
Bain Capital Private Credit - $350mm notes due 2031 at T+290, no tightening from talk
Cox Capital tenders - Up to $10mm of Ares Strategic Income at 13.4% discount and $15mm of Apollo Debt Solutions at 14.6%, expiring Nov. 14
The Reality Check
Carlyle’s $1 trillion figure deserves to be taken literally. The industry is being asked to finance more than half its current AUM into a sector where the ultimate counterparties number fewer than ten and the revenue model is unproven. That’s the software trade of 2021 with larger checks, harder assets that depreciate faster, and leases that only work if frontier labs keep paying. Jenkins saying it publicly from a firm that still wants the exposure is more credible than a short seller saying it. The 8-10% limits his investors are asking for will become standard, and the managers that already exceed them will spend the next two years explaining why.
Apollo’s daily marks and the SEC’s reminder landed within 72 hours of each other, which tells you the regulator and the largest manager reached the same conclusion. Quarterly marks on an asset class with $15bn of trapped redemptions and a Loparex every month aren’t sustainable. The complaint from rivals is about volatility. The real exposure is divergence. Once Apollo prints a daily price on a loan, every other holder of that loan is marking against a visible number.
OTIC at 39% is the clearest evidence yet that this cycle is sorting by product rather than by credit. The fund’s non-accruals are fine. The fund is still in effective liquidation because a single-sector evergreen vehicle sold to wealth clients has no answer when the sector reprices. Metrics in Australia is the same lesson with a worse structure, monthly liquidity against illiquid loans and a bank facility that may not renew. The US 5% gate looks like discipline by comparison. It’s also only as good as the revolver behind it.
Audax raising $10bn and Oaktree $2bn in the same week Bain couldn’t tighten a bond says the institutional market is open and selective. Industrial middle-market lending with low software exposure clears. Unrated ABF clears. Retail evergreens with 39% queues and BDC bonds at T+290 do not. That’s a functioning market pricing risk, which is healthier than the one that existed in 2024. It’s just a smaller one.