In the summer, Nvidia came up with an idea for making money not only from selling chips, but also from the revenues of companies that rent them out. After roughly three months, however, it is revamping the program, called AI Compute Partnership. According to The Information, which cited participants in the negotiations, the chipmaker is fundamentally reconsidering how the program operates. It is therefore revising some preliminary agreements as well as contracts that have already been finalized.
Double earnings from a single chip
Nvidia unveiled the program on July 1. It was intended for selected AI cloud operators, companies that build data centers filled with graphics chips and then rent out the computing power to other customers. Nvidia was to help them secure financing for hardware purchases and, in return, receive a share of the rental revenue. The company described the program as combining credit support with revenue sharing. In practice, it would thus collect money twice: first for the chips sold and then from the cloud capacity it helped build.
The result was intended to be greater computing power for AI startups, language model developers, large enterprises, research institutions, and government organizations. Nvidia named Sharon AI and Firmus among its first partners. Sharon AI planned to deploy up to 40,000 Grace Blackwell GB300 chips, while Firmus is building a facility on the Indonesian island of Batam that could expand to as much as 360 megawatts and accommodate approximately 170,000 Nvidia chips.
The scale of the financial sums involved was revealed in a quarterly report. As of July 26, Nvidia had agreements with AI cloud operators worth a total of $36 billion. The contracts typically run for six years, and their value gradually declines depending on how much capacity is consumed by external customers or how much Nvidia itself uses for its own research and development. Cloud companies can also stop providing capacity to Nvidia and sell it to other interested parties at a better price. Nvidia receives a share of the revenue only if predefined conditions are met. In the same report, the company warned that potential changes in market conditions could harm its financial performance.
Antitrust concerns and reluctant cloud providers
The first problems emerged as early as August. The Wall Street Journal reported at the time that Nvidia had put some deals under the program on hold because it feared scrutiny from antitrust regulators. The company would gain considerable influence over how participating cloud operators use their capacity. An Nvidia spokesperson responded that the model unveiled in July remained in place and was gradually evolving due to high demand. However, the spokesperson did not disclose which transactions had been suspended or whether the commercial terms were being rewritten.
Only now is it becoming clear why the program also stalled for commercial reasons. Some established cloud companies, including Nebius, refused to participate. They did not want Nvidia to reduce their future profits, and they are also able to secure financing on their own without its help. The other side also had concerns. According to The Information, Nvidia feared that some companies interested in participating would become too financially dependent on the chipmaker. If their businesses relied on money and guarantees from Nvidia, difficulties at one partner could quickly spread back to it.
Nvidia is therefore not canceling the program, but merely changing its terms. The Information did not specify what new percentage of revenue Nvidia wants to take or which specific contracts are affected by the changes.
