Artificial intelligence startups represent a new, riskier breed of tenants that account for a growing share of demand for data centers and the high-powered chips used inside them. But lenders are balking at providing the billions of dollars these “neoclouds” require to expand their computing capacity.

Now, Nvidia is trying to coax insurers into shouldering losses incurred when AI startups default on loans — an innovative finance structure the chipmaker hopes will uncork a flood of expansion capital to some of its fastest-growing customers.

The AI chipmaking giant is proposing structures in which insurers would cover potential losses on loans made to neoclouds to purchase Nvidia GPUs, the Financial Times reports.

It is increasingly common for the chips themselves to serve as the collateral in such financing agreements. Under the structures now being discussed, insurers would compensate lenders in the event a neocloud defaults and the GPUs securing the loan are unable to be resold for a high enough price to repay the outstanding debt. Such agreements could potentially boost access to capital for some of Nvidia’s riskier customers.

The talks are still at an early stage, and no agreements are currently in place, according to the FT.

That Nvidia is pitching such arrangements reflects the fact that it is becoming harder for a growing share of the firm’s customer base to access the capital needed to purchase its products.

Nvidia rocketed from a niche maker of processors best known for powering video games into one of the world’s most valuable companies thanks to an AI arms race that has driven nearly insatiable demand for its products. And in the early days of the AI boom, nearly all of that demand came from tech giants like Amazon, Microsoft, Google and Meta — investment-grade companies that could largely fund their chip purchases from their own balance sheets.

But over the past 24 months, a growing share of the demand driving Nvidia’s growth has come from a different, riskier class of customer: neoclouds.

Neoclouds represent the fastest-growing segment of the data center market. With major players including CoreWeave, Core Scientific, Nscale and Lambda, these AI-specific cloud providers offer on-demand access to the GPU computing critical for AI. The sector posted 223% year-over-year revenue growth in the fourth quarter of 2025, according to Synergy Research Group.

Unlike traditional hyperscalers, neoclouds almost always need to finance their billions of dollars of GPUs and the data centers to house them, a tall task for young companies with short track records and a nascent, largely unproven business model. Their risk profile has made lenders wary. And this poses a potential threat for Nvidia, whose fastest-growing customers need access to vast amounts of capital to keep buying its chips.

Neoclouds’ ability to finance their digital infrastructure has rested, in part, on the use of chips and data centers as collateral. But unlike data centers that hold their value over time, chips are regarded as short-lived assets that depreciate relatively quickly.

While Nvidia CEO Jensen Huang has argued that chips have long useful lives and should be regarded as an “investable asset class,” skeptics of Big Tech's AI spending often cite their potential rapid depreciation as a flaw in the economic assumptions underpinning the data center boom.

Nvidia is hoping insurers and other financial firms can help soften this depreciation risk for potential lenders, according to the FT. Nvidia has reportedly supplied at least one insurer with information on chip depreciation and how the market is expected to value future computing capacity, and it is working with insurance broker Howden Re to develop a structure for such insurance deals.

The potential size of such transactions would be enormous, the FT reports, beyond the balance sheet capacity of individual insurers. As a result, Nvidia is reportedly exploring structures in which insurers syndicate risk to hedge funds and other alternative investors. Nvidia has also reportedly considered participating in these potential financing consortiums itself.

Should these depreciation insurance deals come to fruition, it would be just the latest example of Nvidia using its financial might to reinforce the economic foundations of the firms driving demand for its products. Nvidia has said it has offered to backstop financing deals intended to unlock $500B in capital from institutional investors like Goldman Sachs and Apollo.

Nvidia has provided massive capital injections to major customers and has been increasingly willing to use its balance sheet and credit rating to facilitate deals between major cloud providers, neoclouds and digital infrastructure providers by providing financial support to multiple parties. Huang says this effort is to help build an “ecosystem” of companies that rely on its products.

Last month, Nvidia facilitated the development of Hut 8’s 350-megawatt data center in Nueces County, Texas, for neocloud Lambda by backstopping a $35B cloud deal between Lambda and Claude-maker Anthropic. The deal has Nvidia assuming all the risk for Lambda’s lease of Hut 8’s data center, a facility that will be filled with Nvidia’s chips.

To some, the increasingly tangled financial bloodlines underpinning GPU demand look like a house of cards. The growing frequency of such deals has raised red flags among some investors, drawing accusations of circular financing and comparisons to the “vendor financing” prevalent in the dot-com bubble of the late 1990s.

Data center industry leaders generally dispute this characterization. But the growing share of digital infrastructure investment tied to neoclouds — as opposed to investment-grade tech giants — is beginning to heighten concerns about the shockwaves that any kind of slowdown in demand for AI computing would send through the economy.

As the risk attached to these upstart companies becomes linked to the industry’s hyperscale stalwarts, institutional investors and debt markets, the potential blast radius is getting larger. The addition of insurance companies into the mix is unlikely to ease skeptics’ anxiety.