When GPU prices start to fall, will the "NVIDIA card collateral loan" model of CoreWeave cool down?

In the past two years, the development of AI has made NVIDIA's chips one of the hottest "fashion items" globally. This phenomenon has also given rise to the "GPU economy" on Wall Street, with a number of startups becoming wealthy by acquiring and mortgaging NVIDIA chips to obtain debt financing.

They are also known as "GPU scalpers," with CoreWeave being the most famous among them. This computing power rental company has raised over 10 billion yuan in debt financing over the past year by mortgaging NVIDIA chips, obtaining funds from financial institutions such as Blackstone and Magnetar Capital, with its company valuation skyrocketing from $2 billion to $19 billion in just 18 months.

However, as the prices of NVIDIA GPUs have significantly declined, the "CoreWeave model" has begun to raise concerns. According to a report by the Financial Times on November 4, the trading prices of some market GPUs have plummeted in recent months, with the current price for one hour of GPU computing power being about $2, far below the $8 seen earlier this year.

Critics question that as new advanced GPU versions are released, or as tech giants' current high spending on artificial intelligence begins to shrink, the ongoing value of these "loan collateral" chips will be called into question.

Nate Koppikar from hedge fund Orso Partners stated:

All the lenders coming in are pushing this story, claiming you can use these chips as collateral for loans, which further fuels the frenzy for GPU loans. But chips are depreciating assets, not appreciating assets.

Mortgage Loans and Hoarding More Chips

Cloud computing and "computing power rental" companies like CoreWeave, Crusoe, and Lambda Labs have acquired tens of thousands of high-performance computing chips from NVIDIA. These chips are crucial for developing generative AI models and are used as collateral for massive loans.

Over the past 12 months, CoreWeave has raised over $10 billion in debt financing from lenders such as Blackstone and hedge fund Magnetar Capital, with these debts secured by the NVIDIA GPU inventory held by the company, and the funds raised are used to purchase more chips. This massive financing also means that CoreWeave's leverage is extremely high.

CoreWeave's peer, Lambda Labs, also secured a $500 million loan from Macquarie in April this year, while Crusoe raised $200 million in debt financing from New York investor Upper 90 last year. Similar to traditional asset-backed loans, in the event of default, lenders will have ownership of the GPUs and contracts signed with companies like CoreWeave (i.e., power purchase agreements).

These companies are highly dependent on NVIDIA, although NVIDIA denies providing any customers with priority access to chips. However, as GPU prices plummet and supply chains improve, AI chips are no longer scarce products, leading to ongoing questions about the future value of NVIDIA chips used as loan collateral.

GPU Prices Plummet: Is "NVIDIA Card Collateralization" Feasible?

In recent months, GPU trading prices in some markets have plummeted, with the current trading price for GPU computing power at about $2 per hour, far below the $8 earlier this year.

At the same time, although demand remains strong, the current AI chip supply chain has improved, and NVIDIA's dominant position is being challenged. Several tech giants are developing their own AI chips, with competitor AMD rapidly releasing its own high-performance GPUs.

An executive from CoreWeave, a large lending institution, remarked:

A year ago, owning a GPU was like having a golden ticket to enter Willy Wonka's factory. The situation is different now.

Moreover, there is widespread pressure on tech giants that have invested billions of dollars in AI infrastructure, as their investments are not matching the returns. In June of this year, Sequoia Capital partner David Cahn stated that there is a $500 billion gap between the revenue expectations implied by tech companies' AI infrastructure investments and the actual revenue growth of the AI ecosystem.

Reports indicate that mortgage contracts signed by companies like CoreWeave and Crusoe with tech giants are set to expire in the coming years, which could lead to a surplus of chips in the market.

Critics question whether the ongoing value of these "loan collateral" chips will be sustained as new advanced GPU versions are released, or as tech giants' current high spending on artificial intelligence begins to contract