Nvidia tokenizes cloud revenue: new model or financial risk?

What happened?

Nvidia has announced a new optional financing model for its AI cloud partners, according to Tom's Hardware and Bloomberg. The company now allows startups and infrastructure providers to acquire Nvidia hardware in exchange for a share of future revenue they generate. Instead of paying upfront, customers receive credit tokens that they redeem for computing capacity, and Nvidia gets a recurring percentage of cloud service sales. This scheme is detailed in a blog post co-written by CFO Colette Kress, described as a 'revenue-sharing and credit-support' model. The first confirmed partners are Sharon AI, which will build a 72 MW data center in Australia with up to 40,000 Grace Blackwell GB300 GPUs, and Firmus Technologies, which will build a 360 MW campus in Batam, Indonesia, with up to 170,000 GPUs. Sharon AI has already disclosed in an 8-K filing on June 12 that the agreement has a six-year term. The model emerges in a context where Nvidia has identified a financing gap: even with long-term customer commitments, traditional lenders are reluctant to finance large infrastructure deployments, leaving small clouds unable to leverage existing demand.

Why is this important?

This move represents a strategic shift for Nvidia. Traditionally, the company sold chips and earned one-time revenue with margins near 75%. Now, in addition to that revenue, it will receive a commission on its customers' operating income, potentially significantly boosting long-term profitability. However, it also introduces risk: if partners fail to maintain high utilization rates, the variable revenue stream shrinks. This occurs in a context where rapid hardware depreciation (Nvidia refreshes its GPUs annually) pressures cloud operators' margins. Additionally, Nvidia has faced criticism for circular investments, such as its $30 billion stake in OpenAI or backing xAI. This new model avoids that criticism by not injecting capital directly, but offering credit backed by future revenue. According to Bloomberg, developers receive credit tokens in exchange for a portion of future sales, but neither Nvidia nor its partners have disclosed the revenue-sharing percentages. This tokenized scheme could become a standard if successful, but could also create conflicts of interest if Nvidia favors its financed partners over other customers.

Consequences for the market

For AI startups, this financing lowers the barrier to entry, allowing them to access cutting-edge hardware without upfront costs. However, they give up a percentage of future revenue, which can affect profitability. For investors, Nvidia diversifies its revenue sources and aligns with customer success, but also assumes credit risk. Analysts note that the model could strain smaller cloud operators, already under pressure from rapid equipment obsolescence. For example, GPU lifespan shortens with each annual generation, forcing clouds to amortize hardware over shorter periods. Additionally, the model could intensify competition between hyperscalers (AWS, Azure, GCP) and smaller clouds, as Nvidia now has a financial incentive for its partners to succeed. Compared to earlier events, such as Cisco's equipment financing program in the 1990s, this model is more aggressive by tying Nvidia's revenue directly to customer performance. Back then, Cisco offered loans to buy equipment but did not participate in recurring revenue.

What should readers know?

• The revenue-sharing percentage has not been publicly disclosed. According to Bloomberg, neither Nvidia nor its partners have revealed exact percentages, creating uncertainty about the true cost for customers.
• Sharon AI already has a revenue-sharing agreement with Digital Alpha for up to $200 million, according to its CY25 earnings report. This means its revenue is committed to multiple parties, potentially limiting financial flexibility.
• The tokenized model could become a standard if successful, but could also create conflicts of interest if Nvidia favors its financed partners over other customers. For example, it might prioritize GPU allocation to partners using its sharing model, disadvantaging those who pay upfront.
• The model comes amid growing demand for AI infrastructure, with data center spending expected to exceed $200 billion by 2027, according to IDC. Nvidia seeks to secure its position as the dominant supplier, but also assumes credit risks it previously did not have.

“Nvidia is doubling down on AI: not only does it sell the shovels, but it also collects a toll for every nugget of gold its customers mine.”