Nvidia has begun turning its AI infrastructure business into a financing operation, offering selected cloud providers credit support and taking a share of the revenue generated by the GPU capacity it helps bring online. The change is more consequential than another partnership announcement: Nvidia is moving from selling systems to helping customers make those systems bankable, then retaining an economic claim on their use.
Network World, drawing on research from SemiAnalysis, framed the shift as a response to AI’s changing bottleneck. Supply of accelerators, racks and networking remains important, but large deployments now also depend on whether operators can raise enough debt and equity to pay for power, buildings and rapidly depreciating compute hardware. Nvidia’s own July 1 announcement confirms the essential point: it has introduced a “revenue-sharing and credit-support” model for AI cloud providers.
Nvidia says the arrangement gives participating AI clouds a way to procure Nvidia infrastructure for startups, enterprises and independent software vendors, while Nvidia earns its normal product revenue plus a share of cloud revenue associated with supported capacity. The company has named Australian neocloud Sharon AI and Firmus Technologies as early participants.
What Nvidia has not disclosed is just as important. It has not published the credit-support commitments, the minimum revenue terms, the duration of its obligations, its revenue-share percentage, the accounting treatment, or the conditions under which it must step in. Those omissions mean this is not yet a transparent financial product that customers, competitors or investors can evaluate from Nvidia’s public announcement alone.
The mechanism is more than Nvidia helping a customer negotiate a loan. Data Center Dynamics reports that Nvidia’s backstop involves agreeing to rent unused GPUs at a fixed rate, in return for a cut of the customer’s cloud revenue. The Information independently reported the same basic structure: Nvidia would promise to rent capacity that a cloud provider cannot place with developers or enterprises.
That changes the economics for lenders. A neocloud operator buying tens of thousands of GB300 GPUs has to persuade creditors that the equipment will produce enough rental revenue over its useful life to cover debt service, power, operations and eventual hardware refreshes. A long-term customer contract helps, but it is not always sufficient when the buyer is a young cloud provider with a short operating record.
An Nvidia commitment to absorb idle capacity can create a floor under projected revenue. That can improve a borrower’s apparent credit quality and reduce the risk premium a lender demands. In plain terms, Nvidia is using its balance sheet and its own demand for compute to help convert hardware orders into financeable infrastructure projects.
The vendor is therefore accepting a role normally played by a mixture of banks, private-credit funds, equipment lessors and large strategic investors. Its reward is two-layered: a hardware sale occurs when the capacity is built, and Nvidia receives a recurring revenue share if that capacity is used.
That structure gives Nvidia a strong incentive to support the most successful operators. It also creates a clear conflict of interest in the ordinary commercial sense: the company selling the GPUs is helping establish the cash-flow floor that helps customers finance buying those same GPUs. That is not evidence of improper accounting or a sham transaction. It does mean hardware demand and credit availability are becoming more tightly linked than a conventional vendor-customer sale.
Sharon AI’s SEC filings show why financing matters. In June, the company announced a $1.6 billion financing package consisting of approximately $900 million in equity and pre-funded warrants alongside $700 million of 4.75% convertible senior notes due in 2032. Sharon AI said the proceeds would support its six-year Nvidia compute collaboration and broader expansion.
The filing also stated that Sharon AI had expanded to 132 MW of total AI factory capacity, with 102 MW contracted to end customers and more than 55,000 Nvidia GPUs expected to be deployed by mid-2027. Those figures demonstrate that the new model is being aimed at operators that already have real capacity plans and financing needs, not merely speculative GPU resellers.
But there is a discrepancy between what the public materials establish and what readers may assume. Sharon AI’s announced $1.6 billion financing preceded Nvidia’s July 1 description of the revenue-sharing and credit-support model. Its filing does not set out the terms of any Nvidia backstop. Nvidia’s blog confirms Sharon AI is among the early companies working under the model, but it does not say that the June financing was enabled by a Nvidia guarantee.
That distinction matters for anyone trying to calculate how much capital Nvidia itself has placed at risk. The public record confirms a strategic financing relationship exists; it does not quantify Nvidia’s contingent exposure.
The July program formalizes a broader version of that strategy for smaller or regional AI-cloud operators. Nvidia is no longer only cultivating a handful of preferred buyers through investments and supply arrangements. It is presenting a commercial template in which it can support capacity deployment, share in usage revenue and potentially use idle capacity itself.
The company’s rationale is straightforward. Nvidia argues that model builders and inference providers need compute before they can secure it through conventional site selection, power procurement, construction and system bring-up. If Nvidia can help supply and finance the factory, it can get its platform installed sooner and keep it in service longer.
For Windows-focused enterprise IT teams, the immediate relevance is indirect but real. Many organizations will consume this capacity through hosted inference services, GPU-as-a-service platforms, Copilot-adjacent workloads, managed AI platforms or sovereign-cloud offerings rather than buying a 40,000-GPU cluster. A provider with financing support may be able to offer capacity sooner, commit to longer reserved-capacity terms, or expand in regions where hyperscaler options are limited.
The catch is that cheaper or more available capacity is not guaranteed. Nvidia has not published any requirement that participating clouds pass financing benefits through to enterprise customers in the form of lower prices, predictable reservation rates, service-level agreements, portability protections or data-residency commitments. The model is designed to accelerate deployment; it is not an enterprise price-control program.
A cloud operator normally bears the most direct pain when expensive GPUs sit idle. Under the reported backstop design, Nvidia may assume some of that exposure by renting unused capacity. In exchange, it participates in the upside when the cloud operator fills the machines with paying workloads. The result is a closer relationship between Nvidia’s financial results and the utilization of infrastructure built around its products.
That could be a sensible hedge for Nvidia. If the company needs compute for internal research, software development, cloud services or partner workloads, renting capacity from supported operators can be more useful than allowing newly installed GPUs to remain dark. It could also help preserve momentum for Nvidia’s software stack, networking products and future accelerator generations.
Yet the strategy makes a downturn harder to compartmentalize. If enterprise AI spending disappoints, neoclouds could face lower utilization at the same time that lenders become less willing to finance expansion. Nvidia would then confront weaker demand for new systems while potentially carrying commitments connected to unused prior-generation capacity. The supplier becomes more exposed to the operating economics of its customers.
This is the core issue behind the “circular financing” criticism applied to parts of the AI infrastructure market. Nvidia is not merely selling a tool to an independently financed buyer; it can invest in, support, or backstop the buyer that creates demand for its tools. The criticism does not prove the demand is artificial. It does require readers to distinguish booked hardware sales from independent, end-customer demand for the compute those systems eventually deliver.
No public Nvidia filing or company announcement has confirmed such an OpenAI guarantee as of August 3. It should therefore be treated as reported negotiations, not a completed transaction or an existing commitment.
Still, the July 1 program makes the broader idea less far-fetched than it would have been a year ago. Nvidia has publicly acknowledged that long-term commitments often do not unlock financing for capital-intensive AI infrastructure, and it has publicly offered a credit-support model to address that problem. The unanswered question is how far up the capital stack Nvidia is prepared to go when projects move from tens of thousands of GPUs to multi-gigawatt data centers.
For customers, the practical consequence is simple: evaluate an AI cloud provider’s capacity claims as both a technical and financial proposition. Ask whether capacity is installed or merely planned, whether reserved GPUs are contractually available, what happens if the provider’s financing changes, where workloads and data can be moved, and whether the provider’s promised scale depends on a vendor backstop that has not been publicly described. Nvidia may make more AI capacity possible, but its financing model also makes the health of that capacity market more dependent on Nvidia than ever.
Nvidia says the arrangement gives participating AI clouds a way to procure Nvidia infrastructure for startups, enterprises and independent software vendors, while Nvidia earns its normal product revenue plus a share of cloud revenue associated with supported capacity. The company has named Australian neocloud Sharon AI and Firmus Technologies as early participants.
What Nvidia has not disclosed is just as important. It has not published the credit-support commitments, the minimum revenue terms, the duration of its obligations, its revenue-share percentage, the accounting treatment, or the conditions under which it must step in. Those omissions mean this is not yet a transparent financial product that customers, competitors or investors can evaluate from Nvidia’s public announcement alone.
A GPU sale is becoming a cash-flow promise
The mechanism is more than Nvidia helping a customer negotiate a loan. Data Center Dynamics reports that Nvidia’s backstop involves agreeing to rent unused GPUs at a fixed rate, in return for a cut of the customer’s cloud revenue. The Information independently reported the same basic structure: Nvidia would promise to rent capacity that a cloud provider cannot place with developers or enterprises.That changes the economics for lenders. A neocloud operator buying tens of thousands of GB300 GPUs has to persuade creditors that the equipment will produce enough rental revenue over its useful life to cover debt service, power, operations and eventual hardware refreshes. A long-term customer contract helps, but it is not always sufficient when the buyer is a young cloud provider with a short operating record.
An Nvidia commitment to absorb idle capacity can create a floor under projected revenue. That can improve a borrower’s apparent credit quality and reduce the risk premium a lender demands. In plain terms, Nvidia is using its balance sheet and its own demand for compute to help convert hardware orders into financeable infrastructure projects.
The vendor is therefore accepting a role normally played by a mixture of banks, private-credit funds, equipment lessors and large strategic investors. Its reward is two-layered: a hardware sale occurs when the capacity is built, and Nvidia receives a recurring revenue share if that capacity is used.
That structure gives Nvidia a strong incentive to support the most successful operators. It also creates a clear conflict of interest in the ordinary commercial sense: the company selling the GPUs is helping establish the cash-flow floor that helps customers finance buying those same GPUs. That is not evidence of improper accounting or a sham transaction. It does mean hardware demand and credit availability are becoming more tightly linked than a conventional vendor-customer sale.
Sharon AI and Firmus show the scale Nvidia is targeting
Nvidia says Sharon AI plans to deploy up to 40,000 Grace Blackwell GB300 GPUs. Firmus is building a DSX AI factory campus in Batam, Indonesia, which Nvidia says could scale to 360 megawatts and up to 170,000 GPUs. These are infrastructure programs on the scale of regional cloud platforms, not a financing convenience for a few racks in a colocation facility.Sharon AI’s SEC filings show why financing matters. In June, the company announced a $1.6 billion financing package consisting of approximately $900 million in equity and pre-funded warrants alongside $700 million of 4.75% convertible senior notes due in 2032. Sharon AI said the proceeds would support its six-year Nvidia compute collaboration and broader expansion.
The filing also stated that Sharon AI had expanded to 132 MW of total AI factory capacity, with 102 MW contracted to end customers and more than 55,000 Nvidia GPUs expected to be deployed by mid-2027. Those figures demonstrate that the new model is being aimed at operators that already have real capacity plans and financing needs, not merely speculative GPU resellers.
But there is a discrepancy between what the public materials establish and what readers may assume. Sharon AI’s announced $1.6 billion financing preceded Nvidia’s July 1 description of the revenue-sharing and credit-support model. Its filing does not set out the terms of any Nvidia backstop. Nvidia’s blog confirms Sharon AI is among the early companies working under the model, but it does not say that the June financing was enabled by a Nvidia guarantee.
That distinction matters for anyone trying to calculate how much capital Nvidia itself has placed at risk. The public record confirms a strategic financing relationship exists; it does not quantify Nvidia’s contingent exposure.
Nvidia has been moving in this direction for months
This is not Nvidia’s first effort to use capital as a lever for infrastructure expansion. In January, Nvidia announced a deeper collaboration with CoreWeave, including a $2 billion investment in CoreWeave common stock. Nvidia also said it would use its financial strength to help accelerate CoreWeave’s procurement of land, power and data-center shells.The July program formalizes a broader version of that strategy for smaller or regional AI-cloud operators. Nvidia is no longer only cultivating a handful of preferred buyers through investments and supply arrangements. It is presenting a commercial template in which it can support capacity deployment, share in usage revenue and potentially use idle capacity itself.
The company’s rationale is straightforward. Nvidia argues that model builders and inference providers need compute before they can secure it through conventional site selection, power procurement, construction and system bring-up. If Nvidia can help supply and finance the factory, it can get its platform installed sooner and keep it in service longer.
For Windows-focused enterprise IT teams, the immediate relevance is indirect but real. Many organizations will consume this capacity through hosted inference services, GPU-as-a-service platforms, Copilot-adjacent workloads, managed AI platforms or sovereign-cloud offerings rather than buying a 40,000-GPU cluster. A provider with financing support may be able to offer capacity sooner, commit to longer reserved-capacity terms, or expand in regions where hyperscaler options are limited.
The catch is that cheaper or more available capacity is not guaranteed. Nvidia has not published any requirement that participating clouds pass financing benefits through to enterprise customers in the form of lower prices, predictable reservation rates, service-level agreements, portability protections or data-residency commitments. The model is designed to accelerate deployment; it is not an enterprise price-control program.
The risk shifts from the cloud operator toward Nvidia
Nvidia’s new role does not erase demand risk. It reallocates part of it.A cloud operator normally bears the most direct pain when expensive GPUs sit idle. Under the reported backstop design, Nvidia may assume some of that exposure by renting unused capacity. In exchange, it participates in the upside when the cloud operator fills the machines with paying workloads. The result is a closer relationship between Nvidia’s financial results and the utilization of infrastructure built around its products.
That could be a sensible hedge for Nvidia. If the company needs compute for internal research, software development, cloud services or partner workloads, renting capacity from supported operators can be more useful than allowing newly installed GPUs to remain dark. It could also help preserve momentum for Nvidia’s software stack, networking products and future accelerator generations.
Yet the strategy makes a downturn harder to compartmentalize. If enterprise AI spending disappoints, neoclouds could face lower utilization at the same time that lenders become less willing to finance expansion. Nvidia would then confront weaker demand for new systems while potentially carrying commitments connected to unused prior-generation capacity. The supplier becomes more exposed to the operating economics of its customers.
This is the core issue behind the “circular financing” criticism applied to parts of the AI infrastructure market. Nvidia is not merely selling a tool to an independently financed buyer; it can invest in, support, or backstop the buyer that creates demand for its tools. The criticism does not prove the demand is artificial. It does require readers to distinguish booked hardware sales from independent, end-customer demand for the compute those systems eventually deliver.
The larger OpenAI report remains unconfirmed
The model also provides context for more ambitious reports about Nvidia’s potential financing role. Reuters reported on July 26 that, according to The Wall Street Journal, Nvidia was in talks to provide roughly $250 billion in guarantees to help OpenAI lease a 10-gigawatt data-center project in southern Ohio. Reuters said it could not independently verify the report.No public Nvidia filing or company announcement has confirmed such an OpenAI guarantee as of August 3. It should therefore be treated as reported negotiations, not a completed transaction or an existing commitment.
Still, the July 1 program makes the broader idea less far-fetched than it would have been a year ago. Nvidia has publicly acknowledged that long-term commitments often do not unlock financing for capital-intensive AI infrastructure, and it has publicly offered a credit-support model to address that problem. The unanswered question is how far up the capital stack Nvidia is prepared to go when projects move from tens of thousands of GPUs to multi-gigawatt data centers.
For customers, the practical consequence is simple: evaluate an AI cloud provider’s capacity claims as both a technical and financial proposition. Ask whether capacity is installed or merely planned, whether reserved GPUs are contractually available, what happens if the provider’s financing changes, where workloads and data can be moved, and whether the provider’s promised scale depends on a vendor backstop that has not been publicly described. Nvidia may make more AI capacity possible, but its financing model also makes the health of that capacity market more dependent on Nvidia than ever.
References
- Primary source: networkworld.com
Published: 2026-08-03T20:40:57+00:00
Loading…
www.networkworld.com - Related coverage: nvidianews.nvidia.com
NVIDIA and IREN Announce Strategic Partnership to Accelerate Deployment of up to 5 Gigawatts of AI Infrastructure | NVIDIA Newsroom
NVIDIA (NASDAQ: NVDA) and IREN Limited (NASDAQ: IREN) (“IREN”) today announced a strategic partnership to accelerate deployment of next-generation AI infrastructure.nvidianews.nvidia.com - Related coverage: blogs.nvidia.com
NVIDIA Unlocks AI Compute at Scale, Inviting Partners to Power the AI Infrastructure Buildout | NVIDIA Blog
NVIDIA is partnering with AI clouds to deploy large‑scale, multi‑tenant AI factories, aligning economics through a revenue-sharing and credit-support model.blogs.nvidia.com - Related coverage: nvidianews.nvidia.com
Ilya Sutskever’s Safe Superintelligence Inc. and NVIDIA Announce Long-Term Strategic Partnership | NVIDIA Newsroom
Safe Superintelligence Inc. (SSI) and NVIDIA today announced a long-term partnership to rapidly accelerate SSI’s strategic growth...nvidianews.nvidia.com - Related coverage: investor.nvidia.com
NVIDIA Corporation - NVIDIA and Emerald AI Join Leading Energy Companies to Pioneer Flexible AI Factories as Grid Assets
Collaboration Combines AI Factory Design, Energy Resources and Flexibility to Speed Time to Power and Support Grid Reliability HOUSTON, March 23, 2026 (GLOBE NEWSWIRE) - CERAWeek 2026 - NVIDIA and Emerald AI today announced that they are working with AES, Constellation, Invenergy, NextEra...investor.nvidia.com - Related coverage: nvidia.com
Secure Multi-Tenant AI: Sovereign Neo-Cloud Infrastructure With Fractional GPU Sharing S82177 | GTC San Jose 2026 | NVIDIA On-Demand
This session examines how DCAI, a sovereign neo-cloud operator, supports secure multi-tenant access to high-performance AI compute on the Gefion supercompuwww.nvidia.com - Related coverage: nvidia.com
- Related coverage: images.nvidia.com
- Related coverage: blogs.nvidia.com
India Fuels Its AI Mission With NVIDIA | NVIDIA Blog
From AI infrastructure leaders to frontier model developers, India is teaming with NVIDIA to drive AI transformation across the nation.blogs.nvidia.com - Related coverage: axios.com
AI debt boom ramps up with Nvidia bond sale
The world's largest and most cash-rich companies find that they too need investors to finance the AI buildout.www.axios.com
- Related coverage: tomshardware.com
Nvidia offers to take a cut of AI cloud revenue on top of hardware sales in new optional financing vehicle — trades tokens for revenue cut | Tom's Hardware
Sharon AI and Firmus are the first partners, with a combined total of up to 210,000 GPUs.www.tomshardware.com - Related coverage: axios.com
Nvidia reignites "circular" AI concerns as it weighs OpenAI financing guarantee
The chip giant appears poised to tighten its ties with OpenAI. - Related coverage: tomshardware.com
Nvidia and Japan unveil world's first national AI infrastructure — Noetra consortium to build a 140MW Rubin AI factory with 27,500 GPUs | Tom's Hardware
The state-backed buildout comprises 382 Vera Rubin NVL72 racks.www.tomshardware.com - Related coverage: techradar.com
Nvidia doubles down on CoreWeave as early Vera Rubin access and billions fuel massive AI factory expansion plans | TechRadar
Vera CPUs introduce a standalone server option within Nvidia’s infrastructure stackwww.techradar.com - Related coverage: windowscentral.com
Microsoft drops $9.7B on AI cloud power to keep up with demand | Windows Central
The $9.7B deal highlights how AI demand is outpacing compute supply — and why Microsoft keeps signing billion‑dollar contracts.www.windowscentral.com - Related coverage: investing.com
Nvidia in talks with OpenAI to guarantee $250 billion financing for data center, WSJ reports By Reuters
Nvidia in talks with OpenAI to guarantee $250 billion financing for data center, WSJ reportswww.investing.com - Related coverage: economia.uol.com.br
Nvidia negocia garantia de US$ 250 bilhões para data center da OpenAI
A Nvidia está em negociações para fornecer cerca de US$ 250 bilhões em garantias de financiamento para a OpenAI como parte de um projeto massivo de data center, informou o Wall Street Journal.O apoio da Nvidia ajudaria a empresa do ChatGPT a arrendar ueconomia.uol.com.br - Related coverage: datacenterdynamics.com
Nvidia acts as backstop for customer GPUs in return for cut of cloud revenue - DCD
Firmus and Sharon AI are first adopterswww.datacenterdynamics.com - Related coverage: theinformation.com
Nvidia Says It Will Take a Cut of Some Customers’ Cloud Revenues — The Information
Nvidia is using its powerful balance sheet to help more companies buy its expensive AI chips, and possibly make some money back too. It’s promising to financially backstop young cloud providers that rent out its graphics processing units in exchange for a share of their revenues. The backstop ...www.theinformation.com - Related coverage: bruno.digital
Nvidia Launches a Revenue Sharing Financing Model for Neoclouds With Sharon AI and Firmus as First Partners | Bruno Digital
Nvidia will now finance the GPUs it sells and take a recurring cut of the cloud revenue they generate, stacking two income streams per deal. The structure de-risks the buildout for lenders but hardwires Nvidia into the economics of its own customers, and markets noticed.bruno.digital - Related coverage: aiweekly.co
Nvidia Launches Revenue-Share Model With Sharon AI, Firmus | AI Weekly
Nvidia moved a piece of its cloud strategy into the open this week, publishing details of a "revenue-sharing and credit-support" arrangement that lets…
aiweekly.co
- Related coverage: vff.ai
NVIDIA Opens Compute Access via Revenue-Share Model | VFF - The signal in the noise
NVIDIA is introducing a revenue-sharing partnership model that allows AI cloud providers to procure its infrastructure and resell services to startups, enterprivff.ai - Related coverage: sharkalytics.com
Nvidia Is Testing A Revenue Sharing Model With AI Cloud Startups, Sharon AI and Firmus are The First Big Test - Sharkalytics
Nvidia tests a revenue-sharing AI cloud model with Sharon AI and Firmus, linking GPU deployments to recurring cloud revenue.www.sharkalytics.com - Related coverage: techscurrent.com
NVIDIA AI Cloud Revenue Sharing Turns GPUs Into Recurring Revenue
NVIDIA’s new AI cloud revenue-sharing model lets partners finance large GPU deployments while giving NVIDIA a usage-linked cut of supported cloud revenue.techscurrent.com - Related coverage: thegputrade.com
NVIDIA rolls out revenue-sharing for GB300 clouds
New model lets clouds deploy Grace Blackwell GB300 capacity with credit support in exchange for future revenuethegputrade.com
- Related coverage: cincodias.elpais.com