Nvidia’s new financing strategy fails to materialize

Half a trillion dollars on the table

Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR joined Nvidia to raise 500 billion dollars that would turn computing power into a tradable financial asset. Jensen Huang told CNBC, “This is the first time that technology chips become a type of investable asset.” He added, “These are assets that generate income now. They are productive, long-lived, fungible and flexible.” Larry Fink, CEO of BlackRock, went further, saying, “We are seeing a situation reminiscent of the mortgage-backed securities (MBS) market of the 70s.”

Last year’s Wi-Flash

A year earlier Huang sang a very different tune. When the Blackwell chips began shipping in volume, he told a company conference, “You won’t be able to give away Hopper chips for free.” Those same chips are now being labeled “long-lived” and “income-generating.” In practice, rental prices for older chips are actually rising: Silicon Data forecasts a sustained increase through 2028 and expects utilization to reach 70 %. One cloud provider almost doubled the price to a customer when renewing a contract on B200 chips. Reality, it appears, did not follow the CEO’s forecast.

Where the risk hides

Former hedge-fund manager Mark Rubinstein points to a logical failure: “NBS collapsed when too many mortgages were created.” The AI industry sees a flood of data centers, and Chinese open-source models require less compute, two threats to the assumption of perpetual demand. Above all, the basic question remains: will edge labs such as Anthropic and OpenAI, which drive most of today’s demand, ever become profitable? Without profit on the client side, the asset “generates income” stops generating.

Memoranda of Understanding, not signed contracts

It is important to note that this is still not a closed deal. The arrangements are only memoranda of understanding (MOU). Last year Nvidia signed an MOU for 100 billion dollars to invest in OpenAI, which ultimately did not materialize. The advantage of an MOU is clear: you get a headline, and no one asks questions when the money does not move. Even as an experimental balloon, the move shows pressure to turn chip inventory into a financial instrument before the market changes direction.

What’s missing from the definition of “AI factory”

According to Huang, “Nvidia’s computing is not just a chip.” He notes, “There is also CUDA, which boosts output over time and makes the installed base productive beyond its depreciation period.” Huang talks about “a complete AI factory platform” that elegantly ignores the most expensive physical infrastructure. Yet chips and software alone do not produce compute; you need data centers, electricity, real estate and cooling. Without storage and power supply, the new tradable asset remains largely a PowerPoint presentation.