CoreWeave pioneers "GPU collateral financing," giving AI cloud a significant leverage boost, with a scale exceeding $20 billion

In the context of growing demand for AI computing power, the debt financing model using GPUs as collateral is gaining momentum in the cloud industry.

According to media reports, recently, the London-based cloud startup Fluidstack received approval from lenders such as Macquarie to borrow over $10 billion using NVIDIA GPUs as collateral.

Previously, CoreWeave raised $9.9 billion through GPU collateral financing to purchase chips and lease them to clients like OpenAI, pioneering the financing model of "using GPUs as debt repayment."

Now, CoreWeave is leading several AI cloud computing startups in choosing to finance through high-power chip collateral, with total borrowing exceeding $20 billion.

However, due to the short iteration cycle of NVIDIA GPUs, the rapid depreciation of chips has become a potential risk for this financing model.

Reports indicate that considering this risk, the loan interest rates currently demanded by investors have reached several percentage points, exceeding the requirements of most high-yield bond markets.

Fluidstack Follows Up on GPU Collateral Financing

Fluidstack's borrowing scale exceeds the total GPU collateral debt of $9.9 billion previously raised by CoreWeave. This eight-year-old startup had revenue of only $65 million last year and is expected to exceed $400 million in sales this year.

The company, founded in 2017, primarily provides chip leasing services to clients such as Mistral AI in France and Black Forest Labs in Germany.

According to insiders, Fluidstack has also been in talks with Goldman Sachs and JP Morgan regarding potential debt arrangements. The company is drawing funds from lenders as needed and may not borrow the entire approved amount.

Fluidstack CEO Gary Wu stated that this financing model helps the company quickly accumulate capital while avoiding excessive dilution of equity.

Private Credit Funds Actively Provide Funding

Behind GPU collateral financing is the active participation of private credit funds.

Data from McKinsey shows that over the past decade, the private credit market has grown nearly tenfold to over $2 trillion, with these funds optimistic about the rapid growth of the AI cloud industry and the stability of client contracts.

Kurt Tenenbaum, Senior Managing Director at Blue Owl Capital, predicts that 75% of the funding for AI cloud business construction will be used to purchase GPUs.

Rapid Depreciation of NVIDIA GPUs Becomes a Major Challenge

The core issue facing lenders is how long NVIDIA chips can maintain their value.

NVIDIA has shortened the development cycle for new AI chips in recent years, increasing the risk of existing chips becoming obsolete quickly.

According to CoreWeave's estimates, the effective lifespan of chips is 6 years, while NVIDIA's accelerated new product development cycle may lead to rapid depreciation of chips, posing a potential risk.

Reports indicate that considering this risk, the loan interest rates currently demanded by investors have reached several percentage points, exceeding the requirements of most high-yield bond markets, and strong collateral conditions have been set As a result, some companies are trying to use chips from other manufacturers like AMD as new collateral.

Cloud computing startup TensorWave is actively seeking debt financing secured by AMD chips, which will be one of the first transactions secured by AMD chips.

Other companies are choosing to avoid GPU-secured debt and are looking for cheaper sources of financing.

On Monday, CoreWeave announced the acquisition of data center operator Core Scientific for $9 billion, as part of the company's efforts to reduce borrowing costs.

It is reported that since Core Scientific owns its own data centers, this deal will allow CoreWeave to take advantage of infrastructure financing channels that typically have lower interest rates