NVIDIA’s customers are required to continually invest billions of dollars to build out AI infrastructure. And financial support can come from unexpected places. Sometimes, from NVIDIA itself, The Information reports.
NVIDIA and its largest customers constantly have to find a balance. On the one hand, they support each other in the short term, on the other hand, many hyperscalers intend to compete with NVIDIA in the long term. Microsoft, Meta✴, Google and Amazon (AWS) actively use NVIDIA AI chips, while trying to get rid of this dependence and working on their own solutions: Maia, MTIA, TPU, Trainium and Inferentia. Therefore, NVIDIA invests in the so-called neoclouds, focused on providing AI power. Such companies are to some extent able to become an alternative to hyperscalers and are unlikely to be able to develop products that compete with NVIDIA chips.
These include CoreWeave, which NVIDIA is actively helping to develop. In preparation for the IPO, the company disclosed that about 77% of its revenue in 2024 came from just two clients. The largest of them was Microsoft (62%), which also outpaced all of its closest competitors in terms of NVIDIA accelerator purchases. But CoreWeave’s second-largest client was… NVIDIA itself (15%). In early 2023, when demand for AI accelerators from hyperscalers was fantastic, NVIDIA made a curious move – as part of the so-called Project Osprey, it agreed to spend $1.3 billion over four years to lease its own chips from CoreWeave, while simultaneously investing $100 million in it.
CoreWeave has received hundreds of thousands of state-of-the-art AI accelerators and a contract with NVIDIA until August 2027. NVIDIA is indeed actively using leased accelerators, without being distracted by the deployment of AI data centers. If CoreWeave is successfully listed at a valuation of more than $30 billion, NVIDIA’s upfront equity investment in the company will turn into more than $1 billion in just two years. However, for potential investors, such a partnership is a cause for concern, since contracts with Microsoft and NVIDIA expire in 2027-2029.
Image Source: Kelly Sikkema / Unsplash
Such deals are common in the AI market. “Cyclical” financial agreements show how money flows out of companies and back into them in the industry. SoftBank, which is set to become one of OpenAI’s largest investors, has agreed to pay the company $3 billion for the use of its AI products. Microsoft has invested more than $13 billion in OpenAI, giving it a share of the company’s growing revenue. At the same time, OpenAI pays Microsoft billions of dollars annually for cloud services. Amazon and Google have struck similar deals with Anthropic.
Meanwhile, CoreWeave is fueling its expansion with $8 billion in debt and $15 billion in long-term leases for data centers and offices. At one point, CoreWeave even took out $2.3 billion in debt against NVIDIA accelerators to buy more NVIDIA accelerators. Investors are still unsure how to value the company, since a small number of anchor customers, high growth rates, and large debts are a rare combination.
CoreWeave may also play a role in the development of other companies and projects. In particular, we are talking about OpenAI’s intention to subscribe to CoreWeave’s services – a five-year contract for $11.9 billion assumes receiving a stake in the company. During this period, CoreWeave will be able to receive additional funds annually, which will somewhat reduce its dependence on NVIDIA and Microsoft. However, this contract is only an addition to the contracts with Microsoft and Oracle. In addition, an AI data center is being built for the Stargate project, supported by SoftBank and Oracle. OpenAI told its investors that by 2030, Stargate will provide ¾ of the computing power the company needs.