Technology analyst Ben Thompson has warned that Nvidia's efforts to help customers finance AI data centers may broaden the financial risk of the current infrastructure buildout. His argument compares emerging arrangements with earlier periods when new funding channels expanded capital-intensive networks faster than conventional lenders would support.
Thompson begins with financier Jay Cooke's effort to sell Northern Pacific Railway bonds after institutional investors declined them. Cooke marketed the securities to retail buyers, but the railway's continuing demand for capital and a tightening credit environment contributed to the 1873 collapse of his firm. The resulting panic spread through railway finance. The historical analogy is interpretive, not a prediction that AI investment will follow the same path.
The analysis notes that major technology companies have moved beyond funding data centers solely from cash flow. Oracle, Meta, Alphabet and Amazon issued a combined $80 billion of debt between September and November, then raised $194 billion by July 7, 2026 after $108 billion in all of 2025, according to the article. It also cites Google's June plan to raise $85 billion in equity, including $10 billion from Berkshire Hathaway.
Nvidia's proposed answer brings infrastructure investors including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR into repeatable financing platforms for AI factories. Chief executive Jensen Huang argues that Nvidia-based facilities are revenue-producing, broadly reusable assets whose software improves over time. Nvidia is prepared to support opportunities with residual-value financing of up to 25 percent.
Thompson interprets that backstop as a way to lower customers' capital costs while protecting Nvidia's product margins. Unlike equity, which shares upside and dilutes existing owners without adding repayment obligations, the proposed structures place risk with new capital pools and partly with Nvidia's balance sheet. They may also answer competition from lower-upfront-cost alternatives such as Google's TPUs and Amazon's Trainium.
The concern is that insurance assets, pension funds and other long-duration liabilities could become exposed before AI revenue is sufficient to support the installed capacity. If demand and cash flow expand as expected, the guarantees may cost little and conventional debt markets may reopen. If not, the funding innovation could distribute losses more widely.
This is a financial opinion based on the author's synthesis, not evidence that Nvidia or its partners face imminent failure. Its central question is who bears residual-value and demand risk when chip buyers cannot finance expansion through ordinary cash flow or bonds. Thompson's answer is that Nvidia is finding more investors, but in doing so may move the AI boom into pools of capital generally associated with safety.
