The Wall Street pool to unclog Nvidia

Two of the biggest bottlenecks for the development of artificial intelligence are chips and electricity to power data centers. The third is capital. A lot of capital. With funding sources becoming increasingly disputed and expensive, Nvidia announced an agreement with six of the largest private equity firms on Wall Street to mobilize up to $500 billion in the coming years.

The company signed a memorandum of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. The Wall Street pool will structure financial operations having computational capacity as collateral, facilitating access for Big Tech clients to chips and data centers – and to “attractive rates,” said CEO Jensen Huang.

“These financing platforms will help our clients obtain scarce computational resources on a large scale and build AI factories that will drive all sectors and countries in the AI era,” Huang said. Details are still missing on how credit operations will be carried out, but the idea is to take the weight of financing its clients off Nvidia's shoulders.

For Huang, the agreement will be possible because computational capacity is today “a new class of infrastructure investment.” “It's an important milestone for Nvidia and the AI industry,” Huang said. “We have moved from an era in which companies bought chips and built data centers project by project to one in which AI factories can be financed as productive infrastructure – with replicable platforms, long-term capital, and a diversified client base that uses computational power to generate revenue.”

In an interview on CNBC alongside Huang and other executives from the Wall Street consortium, BlackRock CEO Larry Fink said there is currently “a lot of negativity” surrounding AI investments. But according to him, it is necessary to recognize that the technology “will generate an enormous amount of jobs.” “We need to raise this money as quickly as possible,” Fink said.

The consortium was idealized by Huang – who sought out these six firms. None refused. The $500 billion is a lot of money anywhere in the world – but not so much in the parallel reality of artificial intelligence. The creation of 1 gigawatt of AI computational capacity today costs no less than $50 billion. OpenAI, for example, employs around 2 GW and will need much more to continue advancing.

“The expansion of AI infrastructure is something unprecedented,” said Jim Zelter, president of Apollo. “The expectation is that around $8 trillion will be invested in the coming years, a staggering figure. I see a huge opportunity for private capital to finance part of this amount.” Just the hyperscalers – the large cloud providers for AI – are investing more than $1 trillion per year.

Including all data center infrastructure and electricity, the applied capex could reach $10 trillion by 2030. For Bank of America, the MOU with Wall Street firms comes at a good time for Nvidia. “We await more details, but our initial assessment is positive: the burden falls on the consortium, and not on Nvidia's balance sheet,” said the bank's analysts.

“That seems to represent a change in direction from supplier financing, a strategy that generated criticism about financial circularity.” Nvidia has already made a $100 billion investment in OpenAI, the owner of ChatGPT, and has pledged around $250 billion in guarantees. According to BofA, these values represent 15% of the company's cash flow in the next two years.

For the bank, using computational capacity as a transferable asset between companies should sustain resale prices and chip rentals, keeping depreciation at favorable levels. “Financing, not demand, has been the bottleneck – and all hyperscalers report supply limitations,” said BofA. “This $500 billion pool allows buyers without investment grade – such as new laboratories and clouds – to access GPUs and data centers at attractive rates.”

Fink compared the development of this new market to the emergence of mortgage securitization in the 1970s – and said that “the next chapter of financial engineering” is being born. According to Tito Ávila, founder of LIS Capital, Huang himself said that chips have three essential qualities that make them good collateral: they generate constant revenue, like a rented property; they have a long life, and continue to generate revenue for many years; they are fungible, because any cloud provider can run any of the AI models.

“But Wall Street is still understanding this new asset – and, obviously, looking to create its own vehicles and models to meet the demand of the real economy for more computational capacity,” said the manager. Ávila has a position in Brookfield and said that the company, with extensive experience in utility operations, should perform well structuring financing for AI infrastructure.

The announcement of the funding operation with private equity giants occurred two weeks before the release of quarterly results, on the next 26th. The spreads of the five-year CDS for Nvidia's debt are at 80 points, almost double that of Microsoft and well above Alphabet and Amazon. Boosted by the agreement with Wall Street firms, Nvidia's stock opened the day in high, but closed practically stable. This year, the stock rises 17%. The company is worth $5.3 trillion on the Nasdaq.