Nvidia is organising major investments in AI

On Monday evening, Jensen Huang, CEO of Nvidia, announced, alongside the top executives of Apollo, BlackRock, Goldman Sachs, Blackstone, Brookfield and KKR, an agreement to create financing platforms designed to mobilise, over time, more than $500 billion for AI infrastructure, starting with GPUs (the processors that perform the enormous calculations required by the models in parallel).

The (stated) aim of the operation is twofold. On the one hand, to transform computing power into a new asset class in which investors can invest. Jon Gray, chairman of Blackstone, used the example of aviation: a bank looks at the airline’s accounts but also at the value of the aircraft, which, in the event of bankruptcy, can be resold or leased to another carrier. On the other hand, the system would serve to facilitate the raising of capital for AI laboratories, start-ups and so on. One GW of computing capacity, Huang said, costs between 50 and 60 billion, figures that are not exactly easy to finance. BlackRock CEO Larry Fink has announced his intention to offer future securities to pension funds and those with excessive exposure to equities – essentially, high-credit-quality investments offering long-term returns – whilst Goldman Sachs CEO David Solomon has said he is targeting a credit market backed by Nvidia’s computing power.

And so, whilst on the one hand the various platforms would encourage innovation, on the other they would cement Nvidia’s position, which is already hugely dominant, not least thanks to the creation of an ecosystem in which CUDA software – on which most AI programmes are optimised – only runs on the company’s own chips (effectively tying the entire industry to the company).

In any case, the markets reacted with scepticism and, following the initial reports in the Financial Times, Nvidia’s share price fell by more than 2 per cent. Over the past year, the company has been accused of engaging in circular finance, investing billions in clients such as OpenAI, who have purchased GPUs from Huang’s own company. And although the platforms will attract external capital and each project will be assessed independently, the core technology will remain that of Nvidia (which may offer, on a case-by-case basis and at the end of the contract, support or reimbursement for up to 25 per cent of the residual value of the GPUs). However, Nvidia releases more efficient chips almost every year, to which customers migrate, whilst the older ones are leased out but at ever-lower prices, reducing the revenue with which those who purchased the GPUs would need to repay their loans. The AI race is, in fact, rendering chips ‘obsolete’ more quickly, and Michael Burry – the investor made famous by his bet against subprime mortgages in 2008 – argues that major operators are keeping GPUs on their balance sheets for longer than they actually last, concealing costs totalling $176 billion between 2026 and 2028. On the risks, Solomon admitted: “Will the returns all be adequate? Of course not. There will be winners and losers.”