BDC Software Loans: Ares Capital Marks $6.3B at 94.6% of Cost

Ares Capital, the largest listed business development company, held $6.3 billion of software and services loans at 30 September 2026, marked at 94.6% of what it paid. The rest of its $29.2 billion book is marked slightly above cost. So the entire net discount in the biggest BDC sits in software, the sector AI agents are said to threaten. Across four large BDCs, BDC software loan exposure adds up to $12.7 billion at fair value.

BDC software loan exposure, compiled from the filings

The question this post answers is simple. How much money have listed BDCs lent to software companies, how is that money marked, and is it already going bad? Business development companies are the publicly listed vehicles through which a large share of US private credit reaches mid-sized, sponsor-owned companies. Software has been their favourite borrower for a decade, because recurring revenue looks like a reliable source of interest payments.

The table below compiles the figures from each lender's most recent quarterly report. Ares Capital filed its third quarter 10-Q on 27 October 2026. Blue Owl Capital Corporation and Blackstone Secured Lending report the third quarter on 4 November, and FS KKR on 5 November. For those three, plus Golub Capital BDC, the June quarter is the latest filed data. Mixing periods is a weakness, and the table says which period each row comes from.

Every ratio in the PIK column is our calculation from each lender's statement of operations. Software dollars for Blackstone and Blue Owl are the reported percentage applied to reported fair value. Golub's portfolio total is derived the same way in reverse. None of these are new data, but nobody publishes them side by side, and the side-by-side view is where the pattern shows.

Why these five lenders

Ares, Blue Owl, Blackstone, FS KKR and Golub run the five largest listed BDC portfolios that file quarterly with the SEC. Together they hold about $77 billion of investments at fair value. That is a slice of private credit, not the whole of it. Non-traded BDCs and private funds hold far more, and they disclose less.

How much BDC software loan exposure the filings actually show

The four lenders that report a software line hold $12.7 billion of software at fair value, about 19.3% of their combined $65.7 billion portfolio. That is a large number in absolute terms and a modest one in share terms. It is lower than the roughly 29% that sector-wide studies report, for reasons covered further down.

Ares Capital: the largest single software book

Ares reported software and services at 21.6% of its portfolio at 30 September 2026, down from 23.8% at the end of 2025. It is still the largest industry in the book. Financial services is second at 14.6%, up from 12.4%, which tells you where new money went instead. In dollars, software fell from $7,032.3 million to $6,303.8 million in nine months.

Some of that fall is repayments and sales, and some is markdowns. Amortized cost fell 5.2%, from $7,030.9 million to $6,661.9 million. Fair value fell 10.4%. The gap between those two percentages is the markdown, and it is the most important number in this post.

Golub, Blackstone and Blue Owl

Golub carries the heaviest software weight of the group at 26.0% of fair value, down from 27.2% a year earlier. Moody's moved Golub and Blackstone Secured Lending to a negative outlook in June 2026, citing asset quality pressure from software. Blackstone reported software at 18.9%, down from 20.5%. Blue Owl's "internet software and services" line is 11.6%, but that line is narrow. Its healthcare technology line adds another 5.8%, and much of that is software sold to hospitals and clinics.

Ares Capital's software marks carry the whole discount

Here is the original figure. At 31 December 2025, Ares marked its software book at 100.0% of cost. Nine months later the same category sat at 94.6% of cost, an unrealized loss of $358.1 million. Over the same period the rest of the portfolio, $22.8 billion at cost, moved to 100.4% of cost.

That subtraction is ours, and it rests on a total cost of $29,478 million rounded in the press release. A rounding error of a few million does not change the sign. Software accounts for more than all of Ares's net unrealized loss. The other industries combined offset some of it.

What a 5.4-point markdown means in practice

A fair value mark is the lender's own estimate of what the loan would fetch today. A 94.6 mark does not mean Ares expects to lose 5.4% of principal. It means a buyer would demand a higher yield to take the loan, because the borrower's future looks less certain than it did when the loan was written. In my reading, that is exactly what an AI disruption fear should look like at this stage: repriced uncertainty, not realized loss.

For context, Octus measured the average price of all BDC software loans at 97.8% of par at 30 September 2025, before the 2026 software selloff. Ares now marks its software book well below that level. The direction is clear even if the level is a single lender's judgement.

Non-accruals: stress is real, and not mainly in software yet

A loan goes on non-accrual when the lender stops booking its interest as income because it no longer expects to collect. It is the clearest available sign of a loan that has gone bad. Across these five lenders, non-accruals sit well above the levels of 2024, but the worst numbers are not at the most software-heavy lenders.

FS KKR, which does not break out software in its release, has the worst ratio by far at 7.1% of cost. Ares, with the largest software book, has the lowest at 2.4%, flat on the second quarter. Blackstone's 3.6% at cost came down from 4.7% in the first quarter, when Moody's said three companies, Medallia among them, drove the jump. Medallia is a customer experience software company. One large software default can move a lender's whole ratio.

The sector picture is mixed, and the sources disagree

Octus tracked 47 BDCs with high software exposure and found $3.4 billion of debt on non-accrual in the second quarter, 1.86% of debt at cost, down from 2.16% in the first quarter. KBRA, covering 35 rated BDCs, reported the median non-accrual ratio for listed and other fixed-life BDCs rising to 2.75% of cost from 1.81%. Both can be true, because one measures a total and the other a median, across different samples. I would not build a view on either alone.

PIK income: where BDC software loans pay in paper

Payment-in-kind, or PIK, interest is interest the borrower adds to the loan balance instead of paying in cash. It is common in software lending, where fast-growing companies prefer to keep cash for hiring. It also hides stress well, because a borrower that cannot pay can be moved onto PIK and still show up as performing.

Across the four lenders that report it clearly, PIK interest ran at 6.6% to 8.7% of total investment income in the latest quarter. Ares booked $58 million of PIK interest against $763 million of total investment income in the third quarter, about 7.6%. It also booked $67 million of PIK dividends, mostly from preferred equity, which lifts its total non-cash share to about 16.4%. FS KKR's PIK lines sum to about $42 million of $290 million, near 14.5%, the highest share in the group.

Is PIK rising because of software?

The evidence says not much, so far. Octus found that 12.8% of software loans across 155 BDC portfolios carried a PIK feature at September 2025, and that most of it was negotiated at issuance rather than added after a borrower struggled. S&P Global, reported by Alternative Credit Investor in March 2026, described PIK use across BDCs as broadly stable. My view: the PIK share that was agreed upfront is a pricing choice, and the PIK share that appears later is a warning. Filings rarely separate the two, which is the single most useful disclosure BDCs could add.

Why the filings say 19% and sector studies say 29%

The four filings above put software at 19.3% of combined assets. Octus put software at about 29% of 155 BDC portfolios. S&P Global reported a similar 29% of BDC and interval fund assets. JPMorgan, cited by Octus, estimated 20%. These numbers measure different things, and the gap between them is itself information.

A BDC files each borrower under one industry label of its own choosing. A payroll software company might sit in commercial services. A billing platform for clinics might sit in healthcare. Octus reclassified borrowers by reading business descriptions and counted any company whose main product is delivered as software. That broader count is the right one for an AI disruption question, because an agent that replaces a clinic billing platform does not care what label the lender used.

For your own work, assume the true software share at any given BDC is higher than its filed label. Blue Owl is the clearest case. Its internet software line is 11.6%, its healthcare technology line is 5.8%, and the honest exposure sits somewhere above 17%.

Where the AI risk in BDC software loans actually sits

The AI argument against software borrowers is about revenue. If agents compress seats, or replace a product category outright, a borrower's recurring revenue shrinks and so does its ability to carry debt. The 2026 software equity selloff priced that argument into public stocks months ago. Private loans reprice more slowly, because nobody trades them.

The mechanism matters. A software company that loses 10% of its seats to seat compression can usually still pay interest. Sponsor-backed software companies were typically lent money against recurring revenue, with interest coverage set for steady growth. What the company may not be able to do is refinance on the same terms when the loan matures, because the next lender underwrites the new revenue line, not the old one.

The maturity wall is later than the headlines

Octus found that only 9.7% of BDC software loans mature before 2028, with an average maturity of 4.9 years. Oaktree told Octus the risk was the refinanceability of these loans at maturity. I agree with that framing, and it implies something uncomfortable. The quarterly non-accrual data will look calm for longer than the underlying businesses deserve, because most of these borrowers do not need a new lender until 2029 or 2030.

That is also why the Ares markdown matters more than the Ares non-accrual ratio. Marks can move before a default. They are where a change in expected recovery at maturity shows up first.

Which software borrowers are most exposed

BDC managers told Octus that borrowers with proprietary data, regulated end markets and deeply embedded platforms are the most resilient. Octus singled out single-function, tool-type content software without proprietary data as most at risk. That matches what public companies are now writing in their own filings, which we tracked in SaaS 10-K risk factors on AI agents. I would add one more test: whether the borrower prices per seat. A per-seat business carries the AI risk in its revenue formula. A per-transaction or per-record business carries much less.

Where this analysis is weakest

Four honest problems with the numbers above, in order of how much they should change your reading.

First, the periods are mixed. Only Ares has filed for the September quarter. Blue Owl, Blackstone and FS KKR report on 4 and 5 November, and every row except Ares may look different a week after this post. Golub's fiscal year ends in September, so its next figures arrive later in November.

Second, the markdown finding rests on one lender. Ares is the largest BDC, which makes it the most representative single book, but I have not run the same cost against fair value split for the other four. Their filings allow it, and it is the obvious next step.

Third, filed industry labels understate software, as shown above. The $12.7 billion total is a floor for these four lenders, not an estimate of their true exposure.

The case that software lending is fine

The strongest counter-argument is in the same data. S&P Global found that around 80% of software borrowers it covers grew revenue in 2025, with upgrades outpacing downgrades. Morningstar DBRS said software did not meaningfully drive the rise in non-accruals in the second quarter. Ares told investors in July that its software loans carry loan-to-value ratios in the low 40% range, which leaves a large equity cushion below the lender. A 5.4-point markdown on a loan covered more than twice over by sponsor equity is a pricing event, not a solvency event. If agents erode revenue slowly, sponsors can inject equity and lenders can reprice at maturity. That is how this resolves in most scenarios I can construct, and it is a real possibility that this post is early rather than right.

Frequently asked questions

Where to start if your company's debt sits in a BDC

This post is written for founders and finance leads at sponsor-backed software companies. Your lender's marks on your sector are a preview of the terms you will be offered at refinancing. Two things are worth doing before your next board meeting.

First, find your loan in your lender's schedule of investments. Every BDC lists each borrower, its cost and its fair value in the 10-Q. If your loan is marked below 95, your lender is already pricing in a weaker refinancing than your budget assumes. Then compare your maturity date to the window most BDC software loans fall into, which is after 2028. If yours comes due sooner, you will refinance before the market has settled its view on AI, which is the worst time to do it. For a parallel example of reading a maturity schedule line by line, see the CoreWeave debt maturity ladder. For how private credit funds the other side of the AI trade, see AI debt issuance and private credit.

On 4 and 5 November, read one line in each release: software share at fair value. If Blue Owl and Blackstone report the same drift down that Ares showed, lenders are rationing software credit, and pricing at your refinancing will reflect it.

Related reading