brain/
sourcestock-market

Autoresearch: As of 19 Sep 2026, have software-heavy BDC non-accruals widened vs the book, and is ~85% cov-lite true of private credit?

As of 19 Sep 2026: top-10 BDC non-accruals 3.95% / adj 5.95%; software 26% of pressured FV. Syndicated cov-lite ~90%; private-credit senior cov-lite ~14–21%. No issuer ARR/EBITDA series proving AI caused the marks.

Source

Autoresearch: As of 19 Sep 2026, have software-heavy BDC non-accruals widened vs the book, and is ~85% cov-lite true of private credit?

Generated by /autoresearch on 2026-09-19. Synthesized across 2 rounds from PitchBook LCD recaps, a BIS bulletin, an FSCO N-CSR, and LSTA covenant notes. See Provenance. Treat as raw material — review before promoting into a project or thread. Context: vault/projects/stock-market Research-only. No buy/sell/size.

Summary

Step 6 has better collateral-stress color and still no causal AI→ARR print. PitchBook LCD: ten largest public BDCs’ reported non-accrual debt 3.95% of cost in Q2 (+20 bp), adjusted 5.95%; software was 26% of pressured investments at FV in Q1 (up from 19% YE2025). BIS: BDCs have lent $115B to software (1/5 of lending). The chain’s 85% cov-lite number does not match private-credit senior financings: FSCO N-CSR cites ~90% cov-lite in outstanding syndicated loans vs 14% of private-credit senior financings (YE2025); Proskauer-via-secondary has private-credit cov-lite 4%→21% (2023–25). Step 3 should stay partial. Step 6 stays open.

Findings

Software is over-represented in BDC stress; the AI link is inferred

PitchBook LCD (rising non-accruals): top-10 reported non-accrual 3.95% of debt at cost in Q2, adjusted 5.95%. Across 213 BDCs, adjusted non-accruals jumped to 3.3% in Q1 from a 2.1% three-year average; Medallia and Inovalon were $4.4B of the Q1 total (PitchBook). A second LCD note: 538 of ~5,000 BDC names (10.6%) showed credit pressure at end-March; software was 26% of pressured FV, up from 19% at YE2025; first-lien/unitranche under pressure $35.4B, +44% QoQ (PitchBook). Q1 12-BDC report: software marks falling faster than the rest of the book; software’s share of distressed debt remains in line with its book share (PitchBook report).

BIS Bulletin 128: “They have lent around $115 billion to software firms, which represents about a fifth of all their lending”; spreads “did not rise to reflect the higher cash flow uncertainty from AI disruption risk, and have in fact narrowed” (BIS). CAIA secondary: Golub (~26% software) cut its dividend 15%; BCRED’s Feb 2026 monthly loss cited Medallia; Blue Owl tech vehicles saw 40.7% redemption requests with 0.6% non-accruals — sentiment, not payment default (CAIA). That last pair is exactly Danieli’s suppressed-indicator shape — and it is not an EBITDA/ARR time series.

85% is the syndicated market, not monoline direct lending

FSCO N-CSR (period to 31 Dec 2025): “approximately 90% of outstanding syndicated loans featured covenant lite structures… compared to 14% for private credit senior financings.” Upper-market (EBITDA >$100M) is where PC competes with BSL; below $50M EBITDA, cov-lite is 3% (FSCO N-CSR). A 2026 explainer citing Proskauer: private-credit cov-lite 4% (2023) → 21% (2025) (Angel Investors Network). LSTA’s 1Q26 covenant note names a “Saaspocalypse” as a documentation-market backdrop; the 2Q26 note is about portability, not an 85% PC print (LSTA 1Q26; LSTA 2Q26).

Contradictions and open questions

  • Step 6’s discriminating test (software-heavy vs total non-accruals) is only partly runnable: software is 26% of pressured FV vs ~20% of the book — a modest over-weight, not a blow-out spread, and LCD says software’s share of distressed debt is in line with book share.
  • Medallia/Inovalon dominate dollar non-accruals — two names can move the print without an AI-capability shock.
  • 85% cov-lite as stated for “some of these markets” is true of BSL, false as a PC-senior average. The Swiss-cheese / suppressed-default corollary is weaker on middle-market direct lending (3–14% cov-lite) than on upper-market / BSL lookalikes.
  • No retrieved BDC 10-Q decomposes software-borrower ARR/EBITDA vs AI.

Provenance

Rounds run: 2 of 3 (early-exit — LCD + BIS + FSCO answered the two numeric gaps). X sources: attempted via X search_news; spend-cap 403. Not used. Grokipedia: not used for any 2026 claim.

Web sources:

Generated: 2026-09-19

Referenced by