The AI boom is facing growing capital requirements
The reason for Nvidia’s entry into financing is linked to how much money the development of AI infrastructure consumes today. According to an estimate by S&P Global Ratings, Alphabet, Amazon, Meta, Microsoft, and Oracle are projected to spend approximately $750 billion on capital investments in 2026, which corresponds to about 38% of their combined revenue. These are companies with some of the strongest balance sheets and largest cash flows in the world. Nevertheless, these already high investments are straining their cash reserves, and the cost of new debt financing for tech companies is rising. The problem is even more pronounced for smaller AI infrastructure providers, new cloud companies, and AI model developers, who lack the credit quality of Microsoft or Alphabet but need billions of dollars to expand their services. Nvidia is therefore creating a link between customers who need computing power and financial groups capable of providing long-term capital. Most of the money is not expected to come directly from Nvidia’s balance sheet, but rather from insurance companies, credit investors, infrastructure funds, and other large asset managers. Meanwhile, the size of the entire market continues to grow. In July, Morgan Stanley reported that expected capital expenditures for data centers in 2026 had increased from an initial estimate of approximately $575 billion to about $800 billion in AI capex in 2026 and $1.12 trillion in 2027. It is precisely this growth that explains why financing is becoming just as important an issue for further AI expansion as the availability of the chips themselves. [1]
Nvidia Wants to Turn GPUs into an Asset Suitable for Financing
The new system is designed so that loans and other forms of financing will be secured by Nvidia products, with collateral potentially including not only the GPUs themselves but also other server rack equipment. For selected projects, Nvidia can guarantee up to 25% of the collateral’s value based on its expected residual value. For financial institutions, it is crucial that such hardware has a high purchase price, is in high demand, and in the event of problems for a specific operator can be transferred or sold to another customer. The entire system is therefore based on the assumption that Nvidia’s products will remain in demand, technologically relevant, and sufficiently easy to value throughout the term of the financing. Nvidia itself claims that its computing platform is well-suited for this type of financing due to its broad applicability across models and workloads, the ability to shift capacity between customers, and the extensive CUDA software ecosystem. For Nvidia, such a market could have direct commercial significance. If banks, insurance companies, and asset managers gain greater certainty in financing data centers equipped with their technology, more customers may be able to secure capital to purchase its systems. The chipmaker thus influences not only the supply of computing power but also the availability of the funds needed to purchase it. It is precisely this combination that makes the current announcement a more significant step than a typical partnership between a technology and a financial company. [2]
CoreWeave has already demonstrated that a billion-dollar loan secured by chips can work
Financing based on the value of GPUs is not an entirely new concept, and one of the most striking examples emerged as early as 2023. At that time, CoreWeave, a U.S.-based provider of cloud AI infrastructure, secured $2.3 billion in debt financing, using Nvidia chips as collateral. The credit facility was led by Magnetar Capital and Blackstone, with other capital providers including Coatue, DigitalBridge, BlackRock, PIMCO, and Carlyle. The collateral included Nvidia H100 GPUs, which were in exceptionally high demand at the time, and the funds were to be used to purchase additional graphics processors, expand data centers, and hire employees. During negotiations with creditors, CoreWeave also addressed the depreciation schedule for the hardware and its relationship to loan repayment, which shows that financial institutions began treating advanced GPUs as assets with measurable residual value as early as three years ago. This example is also relevant to Nvidia’s current plan. The difference lies primarily in scale. While CoreWeave secured a single credit line of $2.3 billion, Nvidia is now collaborating with six of the world’s largest asset managers and financial institutions to secure financing totaling more than $500 billion. If this plan is successful, computing-backed financing could become a standard part of building large AI data centers.
Growing financial ties also carry the risk of circular financing
The biggest question for investors is where support for the growth of the AI ecosystem ends and where a situation begins in which Nvidia indirectly supports demand for its own products. The most visible example is, once again, CoreWeave. In January 2026, Nvidia invested $2 billion in the company and became its second-largest shareholder. Even before this investment, it owned approximately 6.3% of CoreWeave, and the new purchase nearly doubled its stake in the company. CoreWeave operates data centers built largely on Nvidia GPUs and subsequently leases this computing capacity to other technology companies. Nvidia also has a $6.3 billion contract with CoreWeave under which it has committed to purchasing cloud capacity that CoreWeave is unable to sell to other customers. Reuters noted back in May that Nvidia’s investments in companies that also purchase its technology are raising concerns among investors about potential circular financing. The new program with Wall Street further intensifies this debate, as Nvidia can now help secure capital for other customers in its ecosystem and, for some projects, guarantee a portion of the value of the financed assets. Such a system could be very powerful for the company’s revenue growth, as it expands the number of entities capable of financing new data centers. At the same time, however, it heightens the importance of the question of how much of future demand for Nvidia’s products will stem from customers’ natural needs and how much will depend on the availability of new capital. The success of the entire model will therefore depend not only on NVIDIA’s technological lead, but also on whether the data centers will generate sufficient revenue to repay the debt and whether the expensive AI hardware will retain its value throughout the financing period. [3]
[1,2,3] Forward-looking statements are based on assumptions and current expectations, which may be inaccurate, or on the current economic environment, which may change. Such statements do not guarantee future results. They involve risks and other uncertainties that are difficult to predict. Actual results may differ materially from those expressed or implied in any forward-looking statements.
Sources:
