Nvidia is testing an AI infrastructure model that adds recurring cloud revenue to its traditional GPU sales business. Instead of only selling hardware to cloud providers, Nvidia will help AI cloud startups access large-scale GPU capacity through revenue-sharing and credit-support arrangements.
Under the model, Nvidia can earn standard product revenue, then collect a share of cloud revenue generated by supported capacity. Nvidia describes the goal as building large-scale, multi-tenant AI factories with aligned economics between Nvidia and AI cloud operators.
Sharon AI and Firmus are the first major test cases. Together, they represent the potential deployment of up to 210,000 Nvidia GPUs.
Can Nvidia turn AI compute demand into a recurring, usage-linked revenue stream while helping capital-constrained AI cloud startups build massive infrastructure?
Why Nvidia Is Changing the Model
AI demand is shifting toward production inference. Companies need AI factories that can run continuously, generate tokens at scale, and support commercial AI services.
Nvidia says that shift requires large-scale, multi-tenant accelerated computing that can come online quickly, stay highly utilized, and support token-scale AI economics.
Emerging AI cloud companies often have customer demand but lack enough capital to build large infrastructure. Even signed long-term customer commitments have not always convinced lenders to fund major AI cloud deployments.
Nvidia’s model is designed to address that financing bottleneck. AI cloud providers gain access to Nvidia infrastructure. End customers gain more compute capacity. Nvidia earns hardware revenue and potential recurring income tied to cloud usage.
Investor scrutiny adds pressure. It has faced questions over vendor-supported AI financing and circular financing arrangements. A model that helps customers buy Nvidia hardware while giving Nvidia a claim on future revenue will likely draw close attention.
How the Revenue-Sharing Model Works
AI cloud companies procure Nvidia infrastructure and sell Nvidia-powered cloud services to startups, model builders, enterprises, ISVs, researchers, and regional AI players.
Nvidia earns money through two layers:
- standard product revenue tied to infrastructure sales
- a percentage of cloud revenue generated by supported capacity
Nvidia calls the second layer a recurring, usage-linked earnings stream. Tom’s Hardware described the model as Nvidia taking a cut of AI cloud revenue on top of hardware sales.
Nvidia and its partners have not disclosed the exact revenue-sharing percentage.
Reports also say developers may receive token credits in exchange for a slice of future sales. That makes the model a financing vehicle for AI clouds that need large capacity but do not have hyperscaler-level balance sheets.
Participation appears optional and aimed mainly at AI clouds that need infrastructure scale but lack easy access to traditional financing.
Sharon AI – First Test Case in Australia
Sharon AI is one of the first companies working with Nvidia under the new model.
It recently signed a six-year AI infrastructure compute collaboration with Nvidia. The agreement enables 72 MW of new data center capacity in Australia and uses Nvidia’s DSX AI factory design.
Sharon AI plans to deploy up to 40,000 Nvidia Grace Blackwell GB300 GPUs. Planned customers include AI startups, enterprises, university researchers, AI-native companies, government users, research organizations, and hyperscale customers.
Several figures define the Sharon AI deal:
- 72 MW of new Australian data center capacity enabled through the collaboration
- 132 MW of total AI factory capacity after the expansion
- 102 MW already contracted to end customers
- more than 55,000 total Nvidia GPUs expected by mid-2027
- up to $200 million available through a separate revenue-share facility with Digital Alpha
Sharon AI positions the deal as part of its plan to deliver sovereign, large-scale AI compute infrastructure.
Main test for Sharon AI is capital efficiency. The model may help a smaller AI cloud provider scale into a serious infrastructure player, but success depends on high utilization, strong demand, and enough margin after revenue-share obligations.
Firmus – Bigger Regional Bet in Indonesia
View this post on Instagram
Firmus is another early named partner in Nvidia’s new business model.
The company is building a DSX AI factory campus in Batam, Indonesia. The campus is expected to scale to 360 MW and up to 170,000 Nvidia GPUs.
Firmus says AI-native companies need scalable, energy-efficient, and cost-efficient compute infrastructure to compete globally. Its Batam project makes Nvidia’s model a major Asia-Pacific infrastructure test.
Tom’s Hardware identified Firmus as Singapore-based Firmus Technologies and described it as one of the first named partners in the revenue-sharing model.
Key scale markers make Firmus the larger test case:
- 360 MW planned AI factory campus in Batam
- up to 170,000 Nvidia GPUs
- partnership expected to run until 2034
- $25 billion to $30 billion of potential income tied to committed offtake deals during the first six years
- up to 210,000 Nvidia GPUs in combined potential capacity when paired with Sharon AI
Firmus tests a larger question than Sharon AI: can Nvidia’s model support massive regional AI cloud infrastructure outside the biggest U.S. hyperscalers?
Why It Matters for Nvidia
Nvidia’s core business is still tied to selling chips, systems, and full-stack infrastructure upfront. Revenue-sharing adds a second layer: participation in cloud revenue created by Nvidia-powered infrastructure.
If AI cloud providers keep capacity highly utilized, Nvidia can earn money as workloads run over time. That gives Nvidia exposure to training, post-training, fine-tuning, inference, and agentic AI demand after the hardware sale.
It describes the model as a recurring, usage-linked earnings stream. Yahoo Finance described it as a new revenue-sharing approach with AI cloud providers that creates recurring revenue tied to customer usage.
Market reaction and competition add context:
- Barron’s reported Nvidia shares rose 0.6% to $198.83 in early trading after the announcement.
- Barron’s also noted its stocks were up 5.9% year to date through the prior close.
- Alphabet and Amazon are developing or offering access to custom AI chips that compete with Nvidia hardware.
- Prior Nvidia support for neoclouds includes investments and relationships with companies such as CoreWeave and Nebius.
Revenue-sharing with AI cloud providers could help Nvidia broaden its customer base outside major hyperscalers. Sharon AI and Firmus extend Nvidia’s neocloud support by tying the company more directly to cloud revenue generated by deployed capacity.
Why It Matters for AI Cloud Startups
Nvidia’s model could lower the barrier to entry for AI cloud startups that need massive compute capacity but lack hyperscaler-level balance sheets.
Nvidia says the approach can give AI companies faster access to full-stack accelerated computing without waiting through site selection, power procurement, construction, and hardware bring-up.
Target users include:
- startups
- model builders
- enterprises
- research organizations
- regional AI players
- AI-native companies
- inference providers
- agent platforms
- ISVs
Nvidia points to Baseten, Fireworks AI, and Together AI as examples of demand for training, post-training, fine-tuning, and high-volume agentic inference.
For AI cloud startups, the benefit is faster access to infrastructure. A company may be able to support larger customers and production workloads without relying only on traditional lenders.
The tradeoff is future revenue. Startups may pledge a slice of cloud income in exchange for infrastructure access, token credits, or credit support. That could limit upside if demand grows quickly.
Sharon AI shows both sides. Nvidia says the structure gives Sharon AI a more capital-efficient path to scale and lets it support customers that historically lacked access to capital-intensive AI infrastructure. Sharon AI’s separate Digital Alpha revenue-share facility also shows how future income obligations can stack up.
Risks and Criticism
Nvidia’s model creates upside, but it also introduces risk.
Critics may argue Nvidia is double-dipping. Nvidia already earns money through high-margin hardware sales. Taking a downstream cut of cloud revenue adds another claim on customer economics.
Tom’s Hardware framed the model critically, saying Nvidia is reaching into customers’ income statements for additional revenue.
Utilization risk is central. AI cloud operators need expensive GPU capacity to stay busy. If demand disappoints or rental rates fall, partner clouds may struggle to cover hardware costs, operating expenses, and revenue-share obligations.
Hardware refresh cycles add pressure. Nvidia releases new platforms quickly, and newer systems can affect pricing and demand for older capacity.
Investor questions around circular financing may also grow. It has faced scrutiny over financial arrangements involving AI customers, including large-scale participation in OpenAI and xAI-related financing.
Margin pressure could grow if several issues arrive together:
- lower-than-expected utilization
- falling GPU rental prices
- aggressive price competition
- higher power or data center costs
- revenue obligations owed to more than one financing partner
Undisclosed revenue-split terms make the model hard to evaluate. Outsiders cannot yet judge how favorable or risky the structure is for Sharon AI, Firmus, or Nvidia.
Sharon AI’s separate Digital Alpha facility adds another caution point. Multiple financing tools tied to future income can help a company scale faster, but they can also reduce financial flexibility later.
Summary
Sharon AI and Firmus are Nvidia’s first major test of a new AI infrastructure playbook.
Nvidia is trying to convert GPU demand into recurring revenue by helping AI cloud startups finance and deploy large-scale compute, then sharing in the cloud revenue that infrastructure produces.
Upside is a larger role in AI compute economics. Risk is greater exposure to utilization, financing, pricing, and profitability at AI cloud startups buying Nvidia hardware.
Viola Moorhouse is the coauthor and research lead at Sharkalytics.com, specializing in startup performance tracking and investor strategy.
With a background in market research and business journalism, Viola focuses on separating the hype from the reality in the world of televised entrepreneurship. She’s passionate about making complex startup stories accessible to a wide audience.
