The surge in artificial intelligence data center construction is increasingly being financed through private credit and bond markets by hyperscale cloud providers. The massive capital flowing into this sector presents both risks and opportunities for insurers and lending institutions. Insurance brokers have informed CNBC that they are establishing specialized teams and creating custom insurance policies to meet the growing market demand.
AI data centers are subjecting the insurance industry to a significant stress test. Rapid technological advancements and increasingly complex financing structures are posing unique challenges and creating new prospects for the sector. According to McKinsey projections, global data center-related expenditures could reach $7 trillion by 2030. A substantial portion of this investment can no longer be shouldered solely by hyperscale cloud providers. Major technology firms are increasingly turning to private equity, private credit, and bond issuances to fund the construction of these capital-intensive facilities. Data from Preqin indicates that last year, transactions involving private capital in data center infrastructure consistently exceeded $10 billion. The largest single deal amounted to $40 billion, involving a consortium including Nvidia, Microsoft, BlackRock, and Elon Musk's xAI for the acquisition of Aligned Data Centers.
Tom Harper, head of the data center practice at insurance broker Gallagher, told CNBC that the enormous capital invested in data center construction and operation over the past four to five years represents a genuine stress test for major insurance carriers. "When you invest $10 to over $20 billion in a single location, it challenges the market's underwriting capacity. The market has been willing to assume these risks because the projects are built to exceptional standards, incorporate cutting-edge technology, and are located in top-tier infrastructure zones. However, the capacity to provide adequate insurance coverage for these sites has been constrained," Harper stated. Harper noted that in 2023, securing reasonable insurance for a $20 billion campus was nearly impossible. By 2026, discussions about such projects are happening weekly.
Rajat Rana, a partner at Quinn Emanuel Urquhart & Sullivan, remarked, "We are talking about a scale of trillions of dollars, with financing structures that have very little transparency. It feels like a return to previous cycles, but the magnitude is astronomical." Investment related to AI data centers has been described as the largest peacetime investment project in history. Rana further emphasized that it is "the largest peacetime investment project in human history, and it is primarily being conducted through off-balance-sheet financing." Rana, who was involved in litigation related to structured real estate finance after the 2008 financial crisis, said observing the development of AI data center financing evokes a sense of déjà vu.
The AI boom is not only driving demand for data centers but also accelerating advancements in core technologies like power supply and semiconductors. These rapid technological iterations, combined with massive capital inflows, present both risks and potential rewards for insurance providers and lenders.
Customized Insurance Policies Data centers have unique insurance requirements, needing coverage for both real estate and technological assets. Harper mentioned that several major global insurance carriers have established dedicated data center practice groups to manage these complex projects. These facilities carry distinct risks due to their high asset concentration, significant power demands, and use of frontier technology. However, these factors also make them attractive to insurers, often commanding better pricing. Nevertheless, the difficulty of risk dispersion increases significantly when $20 billion in assets is concentrated in areas prone to high winds or hurricanes. Supply chain disruptions add another layer of complexity. High-value equipment frequently imported from overseas is often stored at third-party sites before installation, creating additional risk exposure.
The merger and acquisition surge in the sector is also keeping transactional lawyers busy. Law firm Kirkland & Ellis noted that several firms have formed dedicated data center teams, bringing together experts in real estate, power, telecommunications, finance, insurance, trade, private equity, and cybersecurity.
Professional services firm Marsh has established a specialized digital infrastructure advisory team to handle increasingly complex contract terms. Last year, Marsh also launched a dedicated insurance facility worth €1 billion ($1.2 billion) to cover data center construction in the UK and Europe. Within seven months, the facility's capacity was expanded to a maximum of $2.7 billion. Alex Wolfson, Senior Vice President of Credit Specialty at Marsh's risk division, stated, "Private credit effectively supplements bank financing and supports long-term power purchase agreements for entities other than hyperscale cloud providers." He explained that as data center lending increases, insurers providing coverage for lender default risk are nearing their underwriting limits. Marsh is developing solutions to support lending institutions in this context.
However, Rana cautioned that as financing moves off-balance-sheet, it becomes difficult for insurers to gain a complete picture of the associated risks. He noted that in January, four U.S. senators called for a government investigation into large tech companies' growing reliance on "complex and opaque bond markets to borrow huge sums." In a public letter, they warned that massive debt could lead to "destructive losses" for financial institutions, potentially triggering a broader financial crisis and impacting the economy.
In a report released in March, Rana stated that declining financing transparency creates secondary litigation risks for downstream investors in private credit funds, such as pension funds, insurance companies, and asset managers. These investors may later discover they did not fully understand the concentration risks. He told CNBC that some private equity funds have already consulted him regarding commercial lease and property valuation disputes. Tenants are negotiating lease renewals, while landlords are seeking higher valuations based on increased demand from AI data centers, leading to disagreements.
The "GPU Debt Treadmill" A central debate surrounding potential financing risks involves the mismatch between the lifecycle of GPUs and the long useful life of data center facilities. CoreWeave, a cloud AI compute provider, was the first company to secure a loan using GPUs as collateral, essentially using the value of high-performance chips as security. Last week, the company announced the completion of its first investment-grade GPU-backed financing, totaling $8.5 billion, with its shares rising 12% on the day of the announcement. Data center structures are typically built to last for decades, whereas the average useful life of a GPU is only about seven years.
Rana commented, "Different data centers disclose varying estimates of equipment useful life to investors during financing." He termed this issue the "GPU debt treadmill," a phrase coined by AI commentator Dave Friedberg. "These AI data centers are like running on a treadmill. Even if financing structures include protective covenants and investment-grade counterparty guarantees, the real risk may be whether today's equity problems gradually transform into credit problems tomorrow." "As new generations of chips are released, data centers may be forced to take on more debt to build new infrastructure. This raises a multi-billion dollar question: How fast can they build, and how fast can they secure financing?"
Harper suggested that the financing costs of such projects will likely continue to drive growth in the asset-backed securities market, with further expansion in commercial mortgage-backed securities issuance. For some insurers like Gallagher, the changing industry landscape presents more opportunity than challenge. Harper noted that GPU lifespans are extending. For assets that depreciate more quickly, companies have innovated by creating custom policies that predefine asset valuation methods. "These facilities are enormous. Determining the value of individual components is a logistical nightmare," he added. He also emphasized the replaceable nature of GPUs, stating that operators are already accounting for their shorter lifespan by building more modular facilities.
Marsh's Wolfson concluded, "A core contradiction exists in data center project finance: lenders typically prefer assets with a useful life significantly longer than the loan term. The relatively short lifespan of GPUs challenges this premise." Consequently, lenders are structuring loans more carefully to protect their interests.
Comments