Insurance funds become new financiers of AI computing power: CoreWeave (CRWV.US) and other AI infrastructure financings collectively turn to insurance capital pools

Notice that CoreWeave (CRWV.US) is seeking to tap into insurers' vast investment pools to finance its AI infrastructure expansion. Co-founder Brannin McBee plans to secure investment-grade ratings for more chip-backed debt, and even for the company itself. Speaking about the capital insurers can provide, McBee said, "It's unlimited. It's like all the money in the world sitting in insurance funds." CoreWeave has raised more than $35 billion this year through chip-backed loans, high-yield bonds, convertible bonds and stock offerings. As of the end of June, its debt and lease obligations totaled about $52 billion. An $8.5 billion loan backed by computing infrastructure and a contract with Meta Platforms (META.US) received an investment-grade rating. CoreWeave is seeking to extend this financing approach to debt backed by contracts with lower-rated AI labs. McBee said CoreWeave's computing capacity for 2027 is almost entirely sold out, and the company has also signed contracts extending into 2028 and beyond. The financing center of gravity for AI infrastructure is shifting toward insurance capital pools CoreWeave's demand is not an isolated case, but an industry model that is being rapidly replicated. On August 10 this year, NVIDIA Corporation announced the signing of memoranda of understanding with Apollo, BlackRock, Inc., Blackstone, Brookfield, Goldman Sachs Group, Inc. and KKR, planning to mobilize more than $500 billion in third-party capital to build a financing platform dedicated to computing assets. The main source of funds is precisely insurers' general accounts and long-duration entrusted capital: Apollo owns annuity underwriter Athene, KKR controls Global Atlantic, Brookfield has a wealth solutions division, and PIMCO and BlackRock, Inc. manage enormous insurance accounts. These platforms lend downstream, mostly to emerging cloud vendors such as CoreWeave, Crusoe and Nebius, as well as project companies building data centers. Borrowers use the money to buy NVIDIA Corporation hardware, then repay debt with cash flow generated by computing capacity offtake contracts. An $8.5 billion delayed draw term loan completed by CoreWeave in March this year was placed into a bankruptcy-remote special purpose vehicle (SPV), backed by GPUs and customer contracts, rated A3 by Moody's Corporation, priced at SOFR plus 225 basis points, equivalent to a fixed rate of about 5.9%, and led by Blackstone's credit and insurance business. Meta Platforms' (META.US) Hyperion data center in Louisiana applies the same logic on a larger scale: $27 billion of SPV debt received an S&P A+ rating, with an issuance yield of 6.58%, maturing in 2049. PIMCO reportedly subscribed about $18 billion, and BlackRock, Inc. subscribed more than $3 billion. Meta holds only a 20% equity stake in the joint venture, and this debt is not reflected on its balance sheet. The asset securitization market is expanding in tandem. As of August this year, data center ABS issuance had reached $38.4 billion, up from $24.1 billion a year earlier and about three times the $13.6 billion for all of 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 completed $4.75 billion in financing in July for a Texas hyperscale campus, priced at SOFR plus 340 basis points. From chip collateral to investment-grade bonds The key that allows this debt to enter the of insurance capital is a "credit substitution" in rating logic. Rating agencies are not rating the chips, but the offtaker's payment obligation: as long as the tenant is an investment-grade hyperscaler such as Microsoft Corporation or Meta, and the contract is take-or-pay, the cash flow is treated as quasi-sovereign-level receivables. The risk is that when the offtaker is replaced by an AI lab, this substitution silently fails. OpenAI's annualized recurring revenue is about $25 billion, while its expected cash burn in 2026 is nearly $27 billion. Its cloud commitments to Microsoft Corporation, Oracle Corporation (ORCL.US) and Amazon.com, Inc. have been itemized, totaling nearly $590 billion. Oracle Corporation itself confirms this danger: about half of its $638 billion order backlog is attributable to OpenAI, its fiscal 2026 free cash flow is negative $23.7 billion, and S&P downgraded it to BBB- in July, just one step away from high yield. Another repeatedly discussed variable is depreciation. The economic life of GPUs is generally believed to be only two to three years, while data center buildings are designed to last 20 to 30 years. Michael Burry has built a public short thesis on this, estimating that hyperscalers will understate depreciation by about $176 billion between 2026 and 2028. This lands squarely on critics' concerns: technology risk is entering insurers' balance sheets in the guise of investment-grade fixed income. Once AI infrastructure cash flow falls short of expectations, losses will transmit in reverse along the chain of "AI projects, private credit, insurance companies, annuity holders, state guarantee funds," while Wall Street takes the spread on the way up, and part of the bill on the way down may be borne by state finances.