Analysis
The AI building boom could affect people who never buy a technology stock. Insurers and retirement funds help finance the data centers behind artificial intelligence. If those investments go bad, policyholders and savers could face consequences.
An August 14 research report by Daniel Oliver of Myrmikan Capital raises a further concern: could losses eventually reach taxpayers? It connects AI borrowing with private-equity firms, which buy and manage businesses, and insurers those firms own.
Several parts of that argument are documented. Insurers hold substantial private credit, meaning lending arranged outside public bond markets. Some AI projects rely heavily on borrowed money. State tax credits can also shift part of the cost of insurer failures to the public.
What remains unproven is the size of the connection. The evidence does not establish that life insurers hold more than $1 trillion of AI project debt, or that taxpayers automatically guarantee repayment.
AI borrowing is growing rapidly
An October 5 analysis by Federal Reserve Bank of Kansas City researchers puts borrowing through investment-grade bonds at about $330 billion through mid-2026. These are bonds rated as relatively low credit risk. Issuance by AI-linked companies was about 10 times the full-year 2023 amount.
The most heavily indebted borrowers included special-purpose vehicles: separate companies created to finance particular projects. Debt financed about 90% of their assets. Neoclouds, computing providers focused on AI, were close behind at 85%.
The largest cloud companies, known as hyperscalers, had a lower ratio of about 44%; chipmakers were at 34%. The study includes technology, energy and industrial businesses, so its totals extend beyond spending exclusively on AI.
The practical issue is how much room a borrower has to absorb losses. A project funded mostly with debt has a thinner financial cushion than one with more owners’ capital.
What the $1.2 trillion figure actually measures
The often-cited $1.2 trillion comes from a JPMorgan analysis reported by Bloomberg on October 7, 2025. It counted investment-grade debt at 75 AI-linked companies, including Apple, Oracle and Duke Energy. Together, they represented about 14% of JPMorgan’s high-grade corporate bond index.
That total covers these companies’ outstanding debt. It does not show how much they borrowed specifically for AI chips or data centers, or how much of that debt insurers own.
Myrmikan’s report dates the Bloomberg article to October 7, 2026, although it was published in 2025. Borrowing has grown since then, but that citation error matters when presenting the figure as current evidence.
The borrower also matters. A company with several profitable businesses has more potential sources of repayment than a project relying on one data center’s revenue.
Insurers’ private-credit exposure is real
Federal Reserve Bank of Chicago researchers document $849 billion of private credit held by life insurers in 2024, about 14% of their assets. Insurers owned by private-equity firms drove important parts of that growth. They also gained access to investments issued by businesses within the same corporate group.
There can be a benefit. Some investments help insurers receive money when they need to pay customers. The researchers identify this advantage for indexed annuities, insurance contracts whose credited returns depend partly on a market index.
The $849 billion is real private-credit exposure, but it is not a measure of AI loans. Comparing it with a separate estimate of the private-credit market cannot establish that insurers own 42% of AI-related private credit.
A separate working paper by Amy Huber, Stefan Huber, Bella Shan and Christina Zhu finds that more than 40% of these insurers’ new investments go into structured securities. These investments package financial assets and their payments. More than 60% came from businesses affiliated with the same private-equity owner.
That raises a conflict-of-interest question: the same corporate group can help create an investment and decide to buy it with money supporting insurance promises. Such transactions deserve scrutiny, although common ownership alone does not prove a loan is unsound.
Meta’s Hyperion deal retains corporate exposure
Meta’s Hyperion data-center deal offers a concrete example. Funds managed by Blue Owl own 80% of the Louisiana project’s joint venture; Meta owns 20%. The buildings and long-lived infrastructure were expected to cost about $27 billion.
Meta agreed to rent the facilities, initially for four years with options to extend. It also offered a residual-value guarantee for the first 16 years of operation. This is a limited promise to make a payment tied to the campus’s value if a lease ends or is not renewed and specified conditions are met.
Meta therefore retains some risk. Its guarantee is capped, so it is neither a complete escape from potential losses nor a promise to cover everything.
Meta says PIMCO and other selected bond investors would buy debt financing part of Blue Owl’s contribution. Its announcement does not establish that life insurers ultimately hold most of those bonds.
The SEC interpretation has a limited scope
On July 29, 2026, Securities and Exchange Commission staff agreed with a Latham & Watkins request about certain data-center financing deals. Securities issued under the described structure fall outside the legal definition of an asset-backed security.
Asset-backed securities generally draw repayment from a pool of financial assets, such as mortgages or car loans. The classification matters because it determines whether certain disclosure and risk-retention rules apply. Risk retention requires the party packaging an investment to keep some exposure to losses.
The SEC staff said its letter does not change the law and is not a rule adopted by the Commission. Its conclusion depends on the deal structure described; different facts could produce a different answer.
The letter raises a specific question about investor protections. It does not amount to a blanket repeal of post-2008 safeguards for retirement savings.
Offshore reinsurance complicates the picture
Insurers can transfer risks to other insurers through reinsurance, sometimes across borders. Bank for International Settlements researchers put the reserves associated with risks transferred by major U.S. life insurers at about $2.1 trillion by the end of 2023. Reserves are amounts set aside to meet insurance obligations. The comparable total was about $500 billion in 2017.
Offshore centers took on roughly 40% of transferred risks. Reinsurance can help companies manage risks and capital, but differences in tax rules, asset valuations and required financial cushions can make cross-border arrangements harder to assess.
These figures describe insurance obligations transferred through reinsurance. They do not tell readers how much AI debt backs those obligations.
Guaranty systems can impose public costs
If an insurer fails, a state guaranty association can help protect eligible policyholders. According to the National Association of Insurance Commissioners, that protection has legal limits. Funding comes from charges assessed on other insurers that remain financially sound, with annual limits on those charges.
The system supports covered insurance promises. It does not guarantee unlimited repayment to investors who financed an AI project.
Taxpayers can still bear an indirect cost. The NAIC’s state-by-state chart shows that many states let insurers recover some assessment costs through credits against their premium taxes. These are taxes insurers pay on premium revenue.
When an insurer uses a credit, it pays less tax. The state consequently collects less revenue. Some states spread credits over five years; others use different schedules or provide no equivalent credit.
That can shift costs to public finances, but it does not make taxpayers automatically responsible for failed AI loans. The outcome depends on coverage limits, assessment rules, recovered assets and state tax provisions.
The 2008 comparison needs qualification
Privately valued assets, affiliated transactions, layered financing and concentrated borrowing can make risks hard to trace. Those features justify scrutiny before an economic downturn tests the structures.
The comparison still has limits. A bond repaid from several businesses’ earnings differs from a loan secured by computer hardware that quickly loses value. Buildings and electrical infrastructure may remain useful longer than the chips inside them.
To judge the danger, investors need to know whether borrowers can make payments, refinance debt and absorb falling asset values. They also need to know whether an insurer has concentrated too much money in vulnerable investments.
The warning comes from an investment thesis
Myrmikan describes its research as following Austrian economics, a school of thought that emphasizes markets and the dangers of credit expansion. Its Gold Fund strategy invests in gold-mining shares as protection against a credit collapse. The fund warns that investors could lose substantially all their investment if that event does not occur.
That perspective does not invalidate Oliver’s evidence. Readers should nevertheless distinguish a forecast supporting an investment strategy from a conclusion established by regulators.
The clearest concern is visibility
The central concern is understandable: money supporting insurance and retirement promises helps finance the AI boom. Heavy borrowing and transactions within the same corporate group can increase risks.
What the reviewed evidence does not show is how much AI project debt sits inside life insurers, how badly individual insurers could be hurt, or whether a public rescue would follow.
The next step is to identify where vulnerable loans are concentrated and whether borrowers and insurers can withstand a sharp drop in AI demand. The evidence warrants closer scrutiny. It does not yet establish an inevitable taxpayer bailout.