Author: Dong Jing
Source: Wall Street Journal
NVIDIA is about to start taking a cut from cloud providers' revenue.
NVIDIA is transforming its strong balance sheet into market leverage by providing financial backing to emerging cloud service providers in exchange for revenue sharing, quietly evolving from a chip seller into the "central bank" of the AI computing ecosystem.
On July 1, according to tech media The Information, NVIDIA is providing financial guarantees to young cloud service providers that rent out their GPUs—if these companies cannot find enough AI developers to lease the computing power, NVIDIA will repurchase the unused GPU capacity at an agreed-upon price.
As part of the arrangement, NVIDIA will receive a percentage of revenue from these cloud providers, with the share gradually decreasing over the term of the contract. GPU cloud providers Firmus and Sharon AI have both participated in the project, and three executives with business ties to NVIDIA have confirmed the arrangement.
On July 1, NVIDIA announced on its official website a new business model combining revenue sharing and credit support, enabling AI cloud providers to purchase NVIDIA infrastructure without bearing the full upfront capital expenditure, and to offer computing power services to downstream AI-native enterprises, model developers, and enterprise customers.
Reports indicate that this project is internally referred to by some at NVIDIA as the "AI Compute Partnership." NVIDIA's spokesperson has also confirmed the existence of the project. This move marks a significant strategic shift for NVIDIA:
On one hand, by lowering the financing barriers for emerging cloud service providers, the customer base is expanded; on the other hand, by participating directly in downstream computing power market profits through revenue sharing, control over the AI industry chain is further extended downstream.
Shift in model: from selling chips to sharing cloud revenue
According to NVIDIA's official press release, NVIDIA will receive additional revenue sharing from cloud services, beyond standard product sales, creating a usage-based recurring revenue stream. The core intent of this model is to break down the financing barriers that have long hindered startup AI companies from accessing large-scale computing power.
NVIDIA positions this framework as the "DSX AI Factory" model, targeting AI service scenarios that require continuous cross-regional operations, high utilization, and multi-tenant accelerated computing.
Sharon AI and Firmus are the first cloud providers to participate in this model. Sharon AI plans to deploy up to 40,000 NVIDIA Grace Blackwell GB300 GPUs; Firmus is building a DSX AI factory campus in Batam, Indonesia, with an expected scale of up to 360 MW and up to 170,000 NVIDIA GPUs. These deployments directly reflect NVIDIA’s latest progress in transforming compute demand into tangible, financiable infrastructure.
NVIDIA notes that emerging AI companies have historically faced severe constraints in accessing capital-intensive infrastructure—even with long-term commitment contracts, these often fall short of securing financing for compute procurement. This means that a large number of AI-native companies, model developers, and inference service providers must endure lengthy delays when scaling their computing capabilities: each step—site selection, power procurement, construction, and hardware deployment—can take months or longer.
The promise of the new model is: through realignment of the economic structure, enabling the aforementioned groups to access full-stack accelerated computing capabilities more quickly, without waiting for the traditional infrastructure development cycle to complete.
Fallback logic: Solving the core challenge of GPU financing
According to reports, GPUs are typically the most expensive component in AI data centers. For buyers with lower credit ratings, securing sufficient financing is itself a challenge.
A data center executive commented that NVIDIA's such deals "kill two birds with one stone." He explained that if NVIDIA merely provides backing for the data center lease, "you still face the question of how to finance the GPUs"; but if NVIDIA commits to purchasing unsold computing power within the facility, "the financing issue for the GPUs is solved, and the financing issue for the data center is solved too."
In other words, NVIDIA’s backstop commitment effectively serves as a credit enhancement tool, enabling emerging cloud service providers, who would otherwise struggle to secure bank loans, to access larger amounts of capital and accelerate data center construction.
Strategic intent: Break the monopoly of large clients
NVIDIA's introduction of these initiatives is grounded in a clear strategic context. Currently, a small number of major cloud providers—including Amazon, Microsoft, SpaceX, Oracle, Meta, and Google—have purchased the majority of NVIDIA's chip production capacity. However, several of these companies are developing their own competitive AI chips, posing a potential threat to NVIDIA.
To reduce dependence on these major clients, NVIDIA has spent the past several years supporting a group of emerging GPU cloud service providers, represented by CoreWeave. This "AI Computing Collaboration Program" is a continuation and deepening of this strategy.
According to The Information, NVIDIA has recently been negotiating to provide financial guarantees for OpenAI’s lease of a large data center in Ohio, which, if fully built out at current prices for chips, labor, electricity, and other materials, could cost up to $500 billion.
Capital Investment: From Equity Investment to Production Capacity Guarantee
NVIDIA has made substantial financial investments in this direction.
To date, NVIDIA has invested billions of dollars in equity stakes in several emerging cloud service providers and, in some cases, agreed to lease back their chips, involving companies such as CoreWeave and Lambda, with total transaction values reaching billions of dollars. According to prior reporting by The Information, NVIDIA’s own researchers have also used GPU servers leased back from Lambda.
Regarding capacity guarantees, NVIDIA began advancing related transactions in the fall of last year. In September 2024, NVIDIA committed to purchase all of CoreWeave’s unsold capacity through 2032 if CoreWeave failed to find tenants, at which point the contract was valued at $6.3 billion. This move effectively alleviated investor concerns about CoreWeave’s highly leveraged business model, driving its stock price up nearly 30% in the following week.
According to a regulatory filing by NVIDIA in May of this year (covering the quarter ended in April), NVIDIA subsequently added $3.5 billion to guarantee customer data center leases in exchange for the right to purchase its shares.
Overall, NVIDIA is building a multi-layered alignment mechanism: equity investment, capacity leaseback, lease guarantees, and now revenue sharing. Each layer deepens the financial ties between NVIDIA and downstream cloud service providers, enabling NVIDIA to directly share in the incremental gains from AI computing commercialization beyond chip sales.
Twitter: https://twitter.com/BitpushNewsCN
BitPush Telegram community: https://t.me/BitPushCommunity
BitPush TG subscription: https://t.me/bitpush