The Zhitong Finance App notes that CoreWeave (CRWV.US) is seeking to use insurance companies' huge investment pools to finance its AI infrastructure expansion. Co-founder Branning McBee plans to seek investment grade ratings for more chip-backed debt and even the company itself.
Speaking about the funding that insurance companies can provide, McBee said, “It's unlimited. It's like all the money in the world lying in an insurance fund file.”
CoreWeave has raised more than $35 billion this year through chip mortgages, high-yield bonds, convertible bonds, and stock offerings. As of the end of June, its total debt and lease obligations amounted to approximately $52 billion.
A $8.5 billion loan secured by computing infrastructure and a contract with Meta Platforms (META.US) received an investment grade rating. CoreWeave is looking to extend this funding approach to debt secured by contracts from lower-rated AI labs.
McBee said that almost all of CoreWeave's 2027 computing power production capacity has been sold, and the company has also signed a contract extending to 2028 and beyond.
The focus of financing AI infrastructure is shifting to insurance pools
CoreWeave's appeal is not an exception; it is an industry model that is being rapidly replicated. On August 10 of this year, Nvidia announced that it has signed a memorandum of understanding with Apollo, BlackRock, Bofeng, Goldman Sachs, and KKR to leverage more than 500 billion US dollars of third-party capital to build a dedicated financing platform for computing power assets.
The main sources of funding are insurance companies' ordinary accounts and long-term entrustment funds: Apollo owns annuity underwriter Athena, KKR holds Global Atlantic, Bofeng has a wealth solutions department, and Pinhao and BlackRock manage large-scale insurance accounts.
These platforms lend downstream, mostly to emerging cloud vendors such as CoreWeave, Crusoe, and Nebius, as well as project companies that build data centers. The borrower used the money to purchase Nvidia hardware and then used the cash flow generated from the offtake contract to pay off the debt. A $8.5 billion term loan completed by CoreWeave in March of this year was included in a bankruptcy quarantine special purpose entity (SPV), secured by a GPU and customer contract, and received an A3 rating from Moody's. The pricing was SOFR plus 225 basis points, equivalent to a fixed interest rate of about 5.9%, led by Blackstone's credit and insurance business.
Meta Platforms' (META.US) Hyperion data center in Louisiana applied the same logic to a larger scale: $27 billion of SPV debt received an S&P A+ rating, with an issuance yield of 6.58%. Due in 2049, Pinhao reportedly subscribed for about $18 billion, and BlackRock subscribed for more than $3 billion. Meta only holds 20% of the joint venture, and this debt isn't reflected on its balance sheet.
The asset securitization market is expanding simultaneously. As of August of this year, ABS circulation in data centers has reached US$38.4 billion, up from US$24.1 billion in the same period last year, which is about three times that of US$13.6 billion in 2023. Blue Owl's digital infrastructure credit platform has issued more than $14 billion in data center loans since 2024; Apollo's institutional credit division closed $4.75 billion in July for a Texas hyperscale campus, priced at SOFR plus 340 basis points.
From chip collateral to investment-grade bonds
The key to getting these debts into the eyes of insurance capital is “credit substitution” based on rating logic. The rating agency evaluates not the chip, but the buyer's payment obligation: as long as the tenant is an investment-grade hyperscale manufacturer such as Microsoft or Meta, and the contract is not negotiable, the cash flow is treated as an account receivable at the quasi-sovereign level.
The risk is that when the offender switches to an AI lab, this replacement will silently fail. OpenAI's annual recurring revenue is around $25 billion, and its estimated cash consumption in 2026 is close to $27 billion. Its cloud commitments to Microsoft, Oracle (ORCL.US), and Amazon have been itemized, and the scale is close to $590 billion. Oracle itself has confirmed this risk: about half of its $638 billion order reserve is due to OpenAI, free cash flow for the 2026 fiscal year was negative $23.7 billion, and S&P downgraded its rating to BBB- in July, which is only one step away from high returns.
Another variable that has been discussed over and over is depreciation. GPUs are generally thought to have an economical lifespan of only two to three years, while data center buildings are designed to last 20 to 30 years. Michael Berry has established an open shorting logic in this regard, and estimates that hyperscale manufacturers will depreciate by about $176 billion between 2026 and 2028.
This is on top of critics' concerns: technology risk is entering insurers' balance sheets under the guise of investment-grade fixed income. Once AI infrastructure cash flow falls short of expectations, losses will be transmitted in reverse along the chain of “AI projects, private equity, insurance companies, annuity holders, and state guarantee funds,” while Wall Street takes interest spreads when it rises, and part of the bills when it falls may be borne by the state treasury.