Nvidia turns to insurers to spread the risk of AI build-out

Nvidia has held talks with insurance companies about shouldering the risks of lending against its chips as chief executive Jensen Huang pushes to unlock more demand for its semiconductors beyond Big Tech groups.

The chipmaker has approached insurance companies about a range of structures that could shift some of the risk of capital-intensive semiconductor financing to insurers and other investors, according to people familiar with these discussions.

One idea under discussion is insurance against losses on loans to upstart cloud computing companies, or “neoclouds”, if they default and the Nvidia chips pledged against their debt cannot be resold for enough to repay lenders.

Such protection could encourage more capital to flow to a group of Nvidia customers that lack the balance sheets of Big Tech groups.

The conversations are at an early stage and may not lead to any deals. But the discussions illustrate how the world’s most valuable listed company is experimenting with structures across Wall Street, private capital and now the insurance sector to help expand the range of customers who can buy its chips.

Huang has said chips should be treated like an “investable asset class” akin to other pieces of expensive, long-lasting technology, such as aeroplanes, which support complex financial structures to shift risks and costs between users and investors.

“Nvidia is trying to will the market to participate, to show other capital providers that these are investable assets,” said one person familiar with its efforts to bring in more outside capital.

The talks come as insurers launch a flurry of products aimed at the AI infrastructure build-out. These include coverage for credit risk and falls in chip values, as well as contract breaches caused by power outages or cooling failures at data centres.

The scale of Huang’s ambitions became clearer last month. Nvidia offered to backstop a portion of financing deals intended to unlock $500bn of capital from Wall Street firms such as Goldman Sachs and Apollo.

It also guaranteed $105bn of leases to get a massive data centre built for OpenAI. Nvidia told investors that it expects a quarter of its revenue next year to come from AI labs that the chipmaker supports with its balance sheet.

The insurance discussions represent a new front. These structures could offer protection to the firms that lease chips or lend against them, shifting some of the risk to insurance providers.

Nvidia has shared data on chip depreciation and the expected future price of computing power with at least one insurance firm, the person said. Another person familiar with the talks said Nvidia was working with broker Howden Re on developing a structure involving insurers. Howden declined to comment.

The effort is being led by Ingemar Lanevi, Nvidia’s head of financial solutions, according to people familiar with the discussions. They said Nvidia had explored using insurance groups to syndicate risk to hedge funds and other alternative investors. This is because the potential scale of the deals could overwhelm the balance sheets of even large insurance carriers.

Nvidia has also explored being part of a consortium backing such agreements alongside insurers, hedge funds and asset managers, one of the people added.

Recommended

Nvidia said its “AI infrastructure is an investable asset class because it is uniquely productive, durable and fungible. Leading capital partners help Nvidia and our ecosystem scale financing for the AI factory build-out, expanding compute access for AI-native companies.”

The company is also returning capital on a vast scale, announcing plans to launch a record $150bn share buy-back on Monday.

The structures Nvidia has examined are similar to so-called residual value insurance, which protects against the decline in the value of tech equipment. Companies including Forward Compute and American Compute have sprung up to offer these niche products.

Quentin Saleur, chief executive of Forward Compute, said insurance products could level the playing field for smaller cloud providers by removing the risk that they go bust before fulfilling a contract.

“Large compute buyers don’t want the counterparty risk of a neocloud,” Saleur said. “But when they use insurance, smaller neoclouds are able to compete with an Amazon or a Google, because all of a sudden, the counterparty risk is the same.”

Forward Compute said that it was examining using forward-looking chip valuations from research providers such as Barkr AI and Silicon Data, which have emerged as key data sources for the nascent market.

A forthcoming study from Barkr AI, which provides valuations of assets such as fine art, private jets and AI chips, found that Nvidia’s eight-GPU H100 system from 2022 is worth roughly $320,000 today, about the same as its initial value.

The study projects that if the supply of computing power catches up with demand, the system would retain about two-thirds of its value after one year. After six years, its value would fall to roughly $30,000.

“Nvidia needs valuation data for outside capital to come in,” said Thomas Galbraith, Barkr AI’s founder who is helping bridge the resale data gap. “Lenders want to know what revenue an asset will produce and what you can sell it for over time.”