06-reference/research

naic sapwg private credit ai infrastructure separability

2026-10-23·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
disclosure-surfacesprivate-creditnaic-statutoryai-capexclaim-verification

The adopted SAPWG item never says "private credit," adds no new fields, and the two Schedule BA magnitudes were a real but misdated YE2024 snapshot

The question

Verbatim: "Do the adopted year-end-2026 NAIC SAPWG private-credit reporting changes make AI-infrastructure credit separable in a statutory annual statement, or does it stay buried in Schedule BA — and do the cited Schedule BA magnitudes (6.4% of invested assets, +7.8% YoY) survive a primary-document check?"

Two merged follow-ups from [[2026-08-13-ai-capex-lender-exposure-insurance-pension-vehicles]], both resolving against the same NAIC (National Association of Insurance Commissioners) primary documents in one pass. The parent brief itself flagged the magnitudes as conference-summary-only and barred their reuse until checked. Note at the outset that the question's own framing — "private-credit reporting changes" effective "year-end-2026" — is itself inherited from the secondary source and does not survive contact with the primary text.

What we already know (from the vault)

What the web says

(a) The separability read — the adopted text. The item behind the "private credit reporting changes" characterization is Ref #2025-29, "Reporting Clarifications," exposed December 9, 2025 and adopted by SAPWG on March 23, 2026 (NAIC Ref #2025-29 agenda item).

(b) The magnitude check — against the NAIC primary. Both cited figures trace to real NAIC data, but to year-end 2024, and were already stale when the parent brief used them.

Convergences and contradictions

Synthesis for RDCO

On separability the answer is no, but the honest characterization is "unresolved and trending negative," not "settled." Two things block it today. First, Ref #2025-29 is a bond-definition reporting cleanup: it clarifies the instructions for three elements that already exist, makes no SSAP change, and creates no asset-class category. Second, nothing in the published Schedule BA aggregation carries a sector, industry or collateral dimension, so even a perfect private-credit tag would not separate data-center lending from any other corporate direct lending. But the question is not shut. NAIC staff themselves proposed exactly the right instrument — a new investment characteristic code in electronic column 23 to flag rated notes and rated feeder funds — and the Working Group declined it "at this time," noting it "could be further considered," while routing the related underlying-characteristics question to agenda item 2025-26. That is a live workstream, not a closed door. The defensible claim for RDCO use is: the adopted text creates no separability, the one code that would have was proposed and deferred rather than rejected on the merits, and a successor workstream exists to watch.

The magnitude check is a clean win for the flag discipline, and it should raise our trust in that mechanic. The parent brief refused to let 6.4% and +7.8% into RDCO output pending a primary cite. That refusal was correct: the figures are real NAIC numbers but describe year-end 2024, and the parent presented them as current roughly three months after the NAIC published the year-end-2025 update superseding both. Nothing downstream has to be retracted, because nothing reused them. The correct current pairing is 6.7% and +10.3%, and the more useful fact is the one neither figure captures — Schedule BA is 70% equity, and its clearly credit-shaped portion is somewhere around 1.5% of industry invested assets. Anyone arguing "insurers are loaded with private credit" from a Schedule BA share number is arguing from a schedule that is mostly private equity and hedge funds.

Where this changes the gauge, concretely. The parent brief's recommendation stands: keep the lender leg on data-center ABS new-issue spread, subordination, and BDC non-accruals, which are market prices at daily latency. The statutory addition we flagged is weaker than first written but better-timed. Origination balloon payment percentage is reported on non-financial ABS already, so if we want a stock-of-risk read on how much of that tier is bullet-structured and therefore refinancing-cliff-exposed, it is obtainable from year-end-2025 filings now rather than from year-end-2026 filings in 2027. Three cautions before anyone builds on that. It is an aggregate across all non-financial ABS with no way to isolate AI infrastructure. The year-end-2025 vintage is documented by the NAIC itself as carrying origination-data gaps for secondary-market purchases, so coverage will be uneven until year-end 2026. And the bridge claim "data-center ABS is non-financial ABS" is our inference from the bond-definition categories, not something any of the three NAIC documents states — the words "data center" and "infrastructure" appear zero times across all of them. That inference needs its own verification before it carries weight.

The transferable lesson is about which NAIC document to read, and it applied to us too. Trade and asset-manager recaps of NAIC meetings reliably translate technical blanks items into asset-class narratives, because that is what their readers ask about. The adopted Form A agenda item carries a "Status:" block stating what was adopted and what was declined, and here the declined list was the load-bearing part. Our own first pass then made the mirror-image error in the other direction, reading clarifications of existing columns as new fields — the fix was in a parenthetical ("There are no proposed changes to the applicable categories") in the same document. For any future NAIC claim, the agenda item at content.naic.org/sites/default/files/inline-files/<ref> is the document and the Capital Markets Bureau special reports are the numbers, and both need reading past the headline. Mechanically: naic.org sits behind a Cloudflare challenge that blocks curl, and its PDFs defeat WebFetch's parser, so the working path is WebFetch to land the file on disk, then pdftotext -layout locally.

Why this is in the vault

It retires two specific flagged claims from [[2026-08-13-ai-capex-lender-exposure-insurance-pension-vehicles]] — permanently barring reuse of the 6.4%/+7.8% pairing and replacing it with verified 6.7%/+10.3% — and it closes the statutory leg of the AI-capex lender-visibility question, so the fragility gauge build can stop treating NAIC reporting reform as a pending unlock and can start reading the balloon-payment column that already exists, while knowing its first vintage is incompletely populated.

Open follow-ups

Related

Sources

Vault

NAIC primary (all fetched 2026-10-23)

Secondary — the claim under test, and items NOT verified at primary