Wall Street’s biggest financial institutions have lent more than $11 billion to a group of niche tech companies based on their holdings in Nvidia’s artificial intelligence chips, the world’s most popular product.
Blackstone, Pimco, Carlyle and BlackRock are among the companies that have created a lucrative new bond market over the past year by lending to “neocloud” companies that provide cloud computing to technology groups developing AI products. That’s one.
Neocloud groups including CoreWeave, Crusoe, and Lambda Labs have acquired tens of thousands of Nvidia’s high-performance computer chips, known as GPUs, which are critical to developing generative AI models. These Nvidia chips are now also being used as collateral for huge loans.
The trading frenzy shed light on Silicon Valley’s rampant GPU economy, which is increasingly backed by wealthy New York financiers. But its rapid growth has raised concerns about riskier lending, circular lending, and Nvidia’s potential dominance in the AI market.
The $3 trillion technology group’s chip allocation to Neocloud Group gave confidence to Wall Street financiers to lend companies billions of dollars and use the funds to buy more Nvidia chips. . Nvidia itself is an investor in neocloud companies and is one of the company’s biggest customers.
Critics question the continued value of collateralized chips as new advanced versions come onto the market, or whether current high spending on AI begins to recede.
“Financiers are all pushing the narrative that you can borrow against these chips, adding to the excitement of getting in now,” said Nate Koppiker, a short seller at hedge fund Ortho Partners. . “But chips are an asset that decreases in value, not an asset that increases in value.”
CoreWeave connection
CoreWeave, the largest neo-cloud company based in New Jersey, started accumulating chips in 2017 when it was founded to mine cryptocurrencies, but pivoted to AI two years later. did. The company currently claims to be the largest private operator of Nvidia GPUs in North America, with more than 45,000 chips in stock.
“CoreWeave’s first success was securing GPU capacity from Nvidia at the exact moment ChatGPT and AI were reaching their Cambrian explosion,” said an executive at one of the company’s largest investors.
Backed by venture capitalists and Nvidia, the company’s valuation has soared from $2 billion to $19 billion in the past 18 months. The company is planning an initial public offering (IPO) in the first half of 2025, which could make it even more valuable, people close to the company said.
Coreweave has raised more than $10 billion in debt in the past 12 months from financial institutions including Blackstone, Carlyle and Illinois-based hedge fund Magnetar Capital. This month, the company announced an additional $650 million in financing facilities from Wall Street banks including JPMorgan, Goldman Sachs and Morgan Stanley.
The debt is secured against CoreWeave’s Nvidia GPU stock, and the capital will be used to buy thousands more GPUs. The company plans to have 28 data centers in the US and Asia by the end of 2024, nine times its footprint at the beginning of last year.
This funding means CoreWeave is highly leveraged. When the company announced its initial $2.3 billion debt financing in August 2023, it included about $1 billion in loans from Blackstone, according to two people close to the company. , with annual sales of just $25 million and negative EBITDA of about $8 million. Revenue has soared to about $2 billion this year, one of the people said.
Coreweave declined to comment on its financial details.
Some of CoreWeave’s biggest financial institutions said they were persuaded to invest because of the large deal it negotiated last year with Microsoft, OpenAI’s biggest backer, which could generate more than $1 billion in revenue over several years. This was said by multiple people familiar with the matter.
“The deal with Microsoft was extremely important,” said a person close to the deal. “They won the contract and then said they needed $2 billion of GPUs, and we were able to source it.”
Similar to traditional asset-backed financing, in the event of a default, the lender owns the GPU and the contract with the company that leases the GPU (known as a power purchase agreement).
CoreWeave’s success in raising large amounts of private debt has encouraged more lenders to enter the space. Although most neo-cloud companies are venture-backed, they have unusually large capital expenditures for startups and must rely on credit markets to finance expansion.
Macquarie loaned Lambda Labs $500 million in April, and Crusoe raised $200 million from New York investor Upper90 last year. The Financial Times reported last week that Crusoe was raising $500 million in equity capital from investors including Peter Thiel’s Founders Fund.
In October, Crusoe signed a $3.4 billion deal with alternative asset manager Blue Owl Capital to fund a new data center in Texas that will lease computing power to Oracle and OpenAI.
“Once people start realizing that this is the largest capital investment in human history, the size of the investment starts to make more sense,” said Crusoe CEO Chase Lochmiller.
Exposure to Nvidia
Neoclouds relies heavily on its relationship with Nvidia. For example, CoreWeave has had access to tens of thousands of H100 chips as the chip giant’s “preferred partner,” but its future growth will depend on it gaining access to Nvidia’s new Blackwell chips as well.
Nvidia denies giving priority access to its chips to any customers, including those in which it has invested. “We’re not trying to help someone jump the queue,” Mohamed Siddeek, head of NVidia’s venture capital unit NVentures, told the FT last year.
“From a lender’s perspective, this is a good thing,” said one banker on the debt deal. “They control the entire supply chain. This is a good thing because Nvidia won’t let things get worse.”
However, the prices of GPUs traded in some markets have plummeted in recent months. An hour of GPU computing now costs about $2, down from $8 earlier this year.
Even as demand for chips continues to rise, supply is improving as spare hardware is being resold, and competition for the number of companies building the underlying AI models is narrowing.
Some tech giants are developing their own AI chips, and rivals such as AMD are also racing to release their own high-performance GPUs to challenge Nvidia’s supremacy.
“A year ago, having access to a GPU was like getting a golden ticket to Willy Wonka’s factory,” said a senior executive at one of CoreWeave’s major lenders. “That’s no longer the case.”
Recommended
The future value of Nvidia’s chips used as collateral for loans is also in question. NeoCloud’s lease agreements with technology groups are set to expire in the next few years, likely resulting in a glut of chips available on the market.
Publicly traded technology giants that have spent billions of dollars on AI infrastructure are also under pressure to make big profits. David Kahn, a partner at venture firm Sequoia Capital, said in June that there is a $500 billion gap between the revenue expectations implied by technology companies building AI infrastructure and the actual revenue growth in the AI ecosystem. said.
But neocloud lenders are betting on continued advances in AI.
Eric Falk, Head of Strategy at Magnetar, said, using mobile phone technology as an example, “Forecasting demand is very difficult, but previous forecasts have underestimated future demand in many areas.” ” he said. “The urgency here is immense, and technology companies appear prepared to invest heavily to be at the forefront.”
