AI Data Centers Face Unprecedented Risks as Insurance, Power, and Finance Models Strain

The rapid expansion of artificial intelligence is driving massive investments in data center infrastructure, with individual campuses valued at up to $30 billion. However, the insurance, power supply, and financing models supporting these facilities are facing unprecedented strain and have yet to be tested by a large-scale catastrophic event. The insurance market, particularly the catastrophe bond market which totals around $66 billion, is insufficient to cover the potential losses from a single large data center. While new insurance products are emerging, dedicated data center risk securitization is still 12 to 18 months away.

The second layer of risk involves power supply. Large data centers consume electricity equivalent to a medium-sized city, overwhelming existing grids. This has led to a scramble for power sources, including new gas turbines and long-term power purchase agreements, and even private power plants. Regulatory hurdles, such as New York's year-long freeze on new data center construction, highlight the political and regulatory risks. Companies like Berkshire Hathaway are benefiting by selling power to data centers while also owning construction firms, and nations are competing for these investments, with the UAE investing $46 billion in Germany.

The third, less-discussed layer is financing. Much of the construction is funded by debt secured against long-term leases with a few anchor tenants. This model, common in income-generating real estate, becomes risky due to the concentration of credit risk when a few tech giants act as the sole clients, financiers, and suppliers. An accounting debate over the economic lifespan of AI chips also impacts reported profits. A single physical event at a large campus could simultaneously trigger insurance claims, power supply issues, and tenant payment defaults, a complex risk that models struggle to capture.

Beyond direct physical damage from fires or floods, the most significant risk is revenue loss during downtime. Furthermore, the geographic shift of data centers to regions prone to tornadoes and hail, rather than hurricanes, presents challenges for existing risk models. The increasing classification of data centers as critical infrastructure also brings new risks, including sabotage, war, and cyberattacks, which are particularly relevant in the current geopolitical climate. While institutional investors in Israel are increasing their exposure to alternative assets like catastrophe bonds due to their low correlation with stock markets, this reliance also means that a significant portion of Israeli pension funds, invested in global stock indices, are indirectly exposed to these AI-driven investments. The rapid construction pace outstrips the traditional market's ability to cover these risks, suggesting that the ultimate solution will come from capital markets, but the question remains at what price and whether it will precede or follow a major incident.