A circulating table implies that private equity owners route policyholder money into their own products. The phenomenon is real and large; the table, checked claim by claim against filings and official reports, is wrong in most of its particulars.
| Claims checked against primary sources | Confirmed | Partly | Contradicted | Unverifiable |
|---|---|---|---|---|
| The circulating table, thirteen firms, 104 claims | 46 | 26 | 27 | 5 |
| The video explainer, 30 claims (postscript) | 7 | 18 | 5 | none stated |
What is true
Three official numbers carry the argument and none appears in the table. Ownership is the wrong variable anyway: an insurer needs no private equity owner to hold private credit.
| Measure | Value | Source |
|---|---|---|
| Private-equity-owned US insurers | 137 at year-end 2024, up from 90 in 2018, and 139 by June 2025 | NAIC, repeated by FSOC |
| Their cash and invested assets | $704.3 billion, 7.8 per cent of the roughly $9 trillion US industry; about 96 per cent sits in life companies | NAIC |
| Size of the category over time | Barely existed in 2010; grown by half in six years | NAIC |
| The circulating footnote | ~25 per cent of US insurers PE-backed, which overstates the phenomenon on assets by about a factor of three | The table under review |
| Life insurer investment in private credit | $849 billion in 2024, more than double the 2014 figure | Federal Reserve Bank of Chicago |
| Official risk assessment | Limited exposure to alternative assets and asset-intensive reinsurance means the risk to global financial stability is currently relatively small, while the rapid growth is flagged as something to monitor | IAIS Global Insurance Market Report |
The corrected table
| Firm and insurer | The circulating claim | What the filings say |
|---|---|---|
| KKR / Global Atlantic | AUM grew $72bn to $158bn under KKR | $72bn is the figure at the July 2020 announcement, before KKR owned anything. KKR’s own roll-forward records $97.5bn at closing on 1 February 2021. Growth under ownership is 62 per cent, not 119 |
| Apollo / Athene | ~$274bn, roughly half of Apollo’s AUM | 41.8 per cent ($392.2bn of $938.4bn at year-end 2025); about 40 per cent on mid-2026 figures |
| Brookfield / American Equity | Acquired 2022 | Completed May 2024, valuing the company at about $4.3bn |
| Blackstone / Allstate life unit | Minority stake | A full acquisition for $2.8bn cash, renamed Everlake Life |
| Blackstone / insurance AUM | ~$150bn as of 2021 | Five years stale. About $230bn at Q4 2024 and roughly $271bn since |
| Stone Point / Wilton Re | Owned | Not owned. Canada Pension Plan Investment Board has held 100 per cent since 2014. Stone Point was the seller |
| Eldridge / Security Benefit | ~68 per cent of Eldridge AUM | The denominator is the asset-management holding company formed in December 2024 with about $74bn, not the wider Eldridge group |
| TPG / Hickory Re | $500m plus $150m in TPG stock | The $500m went into Jackson Financial for about 6.5 per cent of its common equity. Hickory Re is Jackson’s own wholly owned captive |
| Footnote | ~25 per cent of US insurers are PE-backed | 7.8 per cent of industry assets, across 137 insurers (NAIC, year-end 2024) |
Carlyle, Ares, Sixth Street, Warburg Pincus and Centerbridge hold up. One row is omitted rather than corrected: it described an unadjudicated federal matter, and no charges have been filed against anyone named. That alone is reason to check the rest.
Where the critique fails
The strongest version of the concern is about governance and measurement. The circulating version is regulatory, and it fails three checks anyone can run.
| The regulatory claim | What the record says |
|---|---|
| Bermuda is an unregulated jurisdiction | Full Solvency II equivalence from the European Commission and NAIC Qualified and Reciprocal Jurisdiction status, which removes the collateral requirement that once applied. Roughly 80 per cent of Bermuda reinsurance business is conducted on a collateralised basis, largely through funds withheld and modified coinsurance, with the assets on the ceding insurer’s own balance sheet. The transparency argument about offshore structures is genuine; the claim that the assets are unsupervised is not supportable. |
| The Federal Reserve has flagged this as systemic | Its Financial Stability Report of May 2026 contains no occurrence of the words “reinsurance” or “Bermuda” in seventy-one pages. Insurers appear there under leverage, illiquid assets and nontraditional liabilities in general terms. |
| Regulators have defined private equity ownership | On the NAIC’s list of regulatory considerations for private-equity-owned insurers, the item titled “Definition of Private Equity” is the one entry with no update in any reporting period. FSOC states the position directly: supervisory focus is on investment behaviour rather than on who owns the company. |
The mechanism
An asset manager acquires a life and annuity company. Those liabilities are long dated and do not redeem like fund capital, so they act as permanent capital: the manager earns fees on assets it would otherwise have to raise, and its origination arm gains a reliable buyer. Two measurement traps follow.
| KKR and Global Atlantic | What the filings show |
|---|---|
| Purchase, 2021 | $4.74bn price, of which KKR’s own cash was $2.91bn; the remainder came from co-investors and rolling shareholders |
| Balance sheet consolidated | $99.5bn of investments against $100.4bn of policy liabilities |
| Remaining stake, January 2024 | 36.7 per cent for about $2.6bn; both tranches priced at one times book value |
| Trap 1: the unit | KKR’s definition of assets under management includes the asset value of the insurance companies themselves, a balance sheet rather than third-party managed money |
| Trap 2: the baseline | Of Global Atlantic’s roughly $213bn of reported AUM at the end of 2025, $58bn is third-party capital in sponsored sidecar vehicles that did not exist at the baseline. Like for like, invested assets went from $144.3bn at the end of 2024 to $158.7bn a year later |
What would tell you something
| The variable to watch | Why it is the one that matters |
|---|---|
| Share of the portfolio that is affiliate-originated, and whether it is disclosed | The NAIC’s work on affiliated-investment reporting is the most consequential regulatory development in this area, precisely because it addresses this variable. |
| Whether capital charges match the risk | Risk-based capital treatment for private credit, asset-backed finance and structured tranches rests on a ratings process designed for corporate bonds. If the charge understates the risk, the capital efficiency is the return. The NAIC’s work on CLO modelling is the live test. |
| Whether liability optionality meets asset illiquidity | Annuities carry surrender features. An illiquidity premium earned against liabilities the holder can exercise is a different object from one earned against liabilities that cannot run. |
| The correlation no single balance sheet shows | If a growing share of the industry holds similar private credit originated by a small number of managers, sector-level correlation is higher than any individual filing shows. Working Paper No. 31 raised this about the high-yield index, where credit risk changed address rather than leaving; it does not become smaller when the buyer is an insurer. |
Postscript, 21 August 2026: the AI channel
A 24-minute video explainer dated 12 August reaches the same readers, claiming Apollo lent Meta $29 billion through Athene with policyholder money, the Big Short, 2026. Thirty of its claims were checked the same way.
| The video’s claim | What the record says |
|---|---|
| Apollo lent Meta $29bn through Athene, funded by policyholders | No Apollo-led Meta financing ever closed. The $29bn figure is the Louisiana (Hyperion) deal. Apollo and KKR bid for it as a team and lost to PIMCO and Blue Owl on 7 August 2025. The only Athene-as-anchor language is a July 2025 newsletter describing “reported” talks. |
| Meta created an SPV and borrowed $30bn privately, lender by lender | A joint venture 80 per cent owned by Blue Owl funds, 20 per cent by Meta; one $27.294bn 144A bond issued by Beignet Investor LLC, 6.581 per cent due 2049, priced by Morgan Stanley on 16 October 2025; PIMCO took about $18bn and BlackRock over $3bn. A “low double-digit” number of investors, insurers among them. |
| If the data center fails, the SPV files and Meta walks away | Meta gave a residual value guarantee for the first sixteen years with a threshold of about $28bn, sized so that sale proceeds plus Meta’s payment make the bonds whole, and discloses a maximum exposure to loss of $46.03bn at 30 June 2026. The leases are four years committed with renewals to twenty. |
| The bonds are A+ because a AA tenant was bundled with a risky asset | S&P’s A+ is one notch below Meta’s own AA-, and is explicitly derived from Meta’s credit through the leases and the guarantee. The notch is a haircut for structure, not a lift. |
| The loans fund the GPUs, priced like real estate | The collateral is land, shell, power and cooling under triple-net leases; the tenant owns the servers. GPU-backed debt is a separate market (CoreWeave’s $8.5bn deal, April 2026). S&P’s data-center securitisation criteria cap ratings in the single-A range because computing technology evolves too fast to project value; Moody’s, Fitch and KBRA methodologies all model obsolescence explicitly. |
| About half of life insurer revenue is investment income | 28 per cent (ACLI, 2024: $336bn of $1.3tn); on the statutory basis about 23 per cent. |
| $800bn, one dollar in five, is in hard-to-sell instruments | $849bn of private placements in 2024 (Chicago Fed), which is 14 per cent of general-account assets, not one in five; one in five appears only if the numerator is divided by bonds alone. |
| CPP committed “well over $10bn” to data centers in a year | Three deals in twelve months is right: C$225m in Cambridge, Ontario (July 2025, a 50 per cent share of a construction loan with Deutsche Bank); a 50/50 European partnership with Goodman with a headline of A$14bn, not US$14bn, and an initial commitment of A$3.9bn (December 2025); US$1.75bn alongside EQT and EdgeConneX (July 2026). CPP’s own money committed is about US$3.2bn, with a ceiling near US$6.6bn if Goodman builds out in full. |
| Life insurers hold about $4tn in bonds | Confirmed: $3.9998tn at year-end 2025 (NAIC), 66 per cent of $6.10tn in cash and invested assets. |
| Health insurers are not carrying this bet | Confirmed: health insurers hold $377bn, 53 per cent in bonds and 21 per cent in cash; life insurers hold 96 per cent of industry mortgage loans and 82 per cent of insurer CLO holdings. |
| Hyperscalers depreciate GPUs over five to six years; Nvidia ships a new architecture every 18 months | Servers and network equipment: Microsoft two to six years, Alphabet six, Amazon revised in 2024 and 2025. No filer discloses a GPU-specific life. Nvidia’s cadence is now annual, stated in its own 10-K. |
| xAI’s Colossus 1 sat idle and was leased to Anthropic and Google | Anthropic: confirmed in SpaceX’s June 2026 prospectus, about 325,000 GPUs across Colossus and Colossus II at $1.25bn a month. Google: a deal exists; its attribution to Colossus 1 is not in the record. “Sat idle” is the video’s inference. |
The channel is real, and it runs through the instruments flagged above: privately placed structured paper and infrastructure debt, from few originators, rated by a process built for corporate bonds. What does not survive is the mechanism the video puts inside the bond. The obsolescent hardware sits in a separate GPU-backed market that Working Paper No. 9 models; the Hyperion notes carry Meta’s credit. The correlation question gains a second face, and the disclosure gap is the same gap.
| The channel, sized | Value |
|---|---|
| US data-centre securitisation, 2025 | About $26bn, more than the previous three years combined |
| Outstanding by mid-2026 | $61bn, against $4bn in 2020 |
| Life companies’ share of insurers’ asset-backed exposure | 79 per cent |
| Privately placed ABS within life insurers’ private placements | About $125bn, tripled since 2017 |
| Buyers of the Hyperion notes | Insurers among them |
| Hyperion tenant commitment | Four years contractual, with the residual value guarantee doing the rest |
| Data-centre breakdown of insurer ABS holdings | None published by the NAIC, just as affiliated-origination reporting is only now being built |
What this note does not claim
Not that private-equity-owned insurers are less safe; that needs comparative default, downgrade or impairment evidence, not presented here. Not that any firm has done wrong; the structure is disclosed and supervised. Not that the phenomenon is systemic; the bodies that would say so have declined to. Not, in the postscript, that the build-out will fail or the Hyperion notes are mispriced; PIMCO sold part of its position above 110 cents within weeks.
The phenomenon is real and quantified; the popular account is unreliable in most of its specifics; what is worth asking about is affiliated-asset disclosure, capital charges and liability optionality, not ownership.
- Method
- Thirteen rows of a circulating table checked claim by claim against primary sources. 104 discrete claims; 46 confirmed, 27 contradicted, 26 partly confirmed, 5 unverifiable. Every correction in section 2 traces to a filing. Postscript (21 August 2026): 30 further claims from a circulating video explainer checked the same way; 7 confirmed, 18 partly, 5 contradicted. Ledger in state/wp35_ai_channel_claims.json.
- Sources
- SEC filings for KKR, Apollo, Blackstone, Brookfield, TPG, Jackson Financial and, for the postscript, Meta Platforms (10-Q, 10-K), Microsoft, Alphabet, Amazon, Nvidia and SpaceX; S&P Global Ratings on Beignet Investor LLC and its data-center securitisation criteria; CREFC, KBRA and SFA issuance data; CPP Investments releases; ACLI Fact Book; NOLHGA; statutory annual statements; NAIC Capital Markets Bureau special report on private-equity-owned insurers; FSOC annual report; IAIS Global Insurance Market Report; Bermuda Monetary Authority; Federal Reserve Financial Stability Report, May 2026; Federal Reserve Bank of Chicago research on life insurer private credit.
- Honesty
- One row of the original table is omitted rather than corrected, because it concerned an unadjudicated federal matter and no charges have been filed against anyone named. This note takes no position on it.
- Related
- Working Paper No. 31 (a safer index is not a safer market), No. 9 (the AI-CDO, which models the GPU-backed tranche the postscript distinguishes), No. 36 (Permissioned Abundance II).