By Stuart Kerr, Technology Correspondent, LiveAIWire
Nvidia announced on July 1, 2026 that it is introducing a revenue-sharing and credit-support model that will allow AI startups, model builders, enterprises, and regional cloud providers to access large-scale GPU infrastructure without the capital commitments that have historically priced out smaller players. The announcement, made in a blog post by Chief Financial Officer Colette Kress and VP Raj Mirpuri, represents a fundamental shift in Nvidia’s business model: from selling hardware outright to taking a recurring share of the cloud revenue generated by the compute it enables. Sharon AI in Australia is already deploying up to 40,000 Nvidia Grace Blackwell GB300 GPUs under the model. Firmus Technologies is building a data centre campus in Batam, Indonesia expected to scale to 360 megawatts and up to 170,000 Nvidia GPUs. The company’s stock stood at $197.58 on the day of the announcement, giving it a market capitalisation of approximately $4.82 trillion.
The move arrives at a moment when Nvidia has already committed more than $53 billion across roughly 170 AI-related investment deals since the start of the AI boom, according to PitchBook data. Its 30 billion dollar investment in OpenAI, announced in late February 2026, was one of the largest AI financing deals ever recorded. Revenue sharing is a different instrument entirely — not equity in specific companies but a structural claim on the usage economics of an entire computing ecosystem that Nvidia is positioning itself to supply. Understanding what this means for startups, for cloud providers, and for the AI industry’s capital structure requires looking at why the model was introduced and what its terms actually do.
Why the Capital Access Problem Exists
The core problem Nvidia’s new model addresses is structural. Large-scale GPU infrastructure requires enormous upfront capital — site selection, power procurement, construction, and hardware procurement that can take years to come online and hundreds of millions of dollars to finance before a single token is generated. Hyperscalers like Microsoft, Google, Meta, and Amazon can fund this from their balance sheets. Microsoft alone spent over 50 billion dollars on data centres in 2025. Startups and regional AI cloud providers cannot. Even when they have long-term customer commitments that should theoretically unlock financing, the capital markets have not consistently treated AI compute contracts as bankable assets in the way that, say, power purchase agreements are bankable for energy infrastructure.
Nvidia’s blog post was direct about this: emerging AI companies historically have had limited access to capital-intensive infrastructure, with even long-term commitments insufficient to unlock financing for compute. The revenue-sharing model is designed to solve this problem by substituting Nvidia’s balance sheet and credit for the startup’s. Instead of a startup needing to raise capital to buy GPUs upfront, it accesses Nvidia-powered infrastructure through an AI cloud partner that has itself accessed that infrastructure under the revenue-sharing arrangement. The startup gets immediate access to full-stack accelerated computing without waiting through the site selection, power procurement, construction, and hardware bring-up cycle.
What the Model Actually Does for Nvidia
The commercial logic from Nvidia’s perspective is equally clear. The revenue-sharing model creates a new recurring, usage-linked earnings stream — described in exactly those terms in Kress and Mirpuri’s blog post — that is a significant evolution from one-time hardware sales. Nvidia currently earns revenue when it sells GPUs. Under the new model, it also earns a share of the cloud revenue generated by those GPUs over their operational life. If the AI cloud services built on Nvidia infrastructure generate substantial revenue — which is the premise of the entire AI infrastructure investment thesis — Nvidia participates in that upside without taking on the operational complexity of running cloud services itself.
The Blockchain.News analysis of the announcement framed this as Nvidia moving from hardware vendor to infrastructure partner with a stake in the outcome — a business model that more closely resembles the royalty model used by semiconductor IP companies like ARM than the traditional chip sale model. For a company that has already captured the dominant position in AI training hardware, extending that position into a recurring share of AI inference economics is the natural next step in the value chain. The AI factories model that Nvidia is promoting — large-scale, multi-tenant, continuously operating compute facilities optimised for token generation at scale — is the infrastructure layer on top of which the entire AI services economy runs. Nvidia is not just selling the picks and shovels. It is negotiating a royalty on the gold.
What This Means for Startups
For AI startups, the revenue-sharing model lowers the barrier to accessing large-scale compute but introduces a new structural cost: a share of revenue flowing to Nvidia in perpetuity, on top of the cloud service costs that would exist in any case. Whether that represents a good deal depends on the counterfactual. A startup that could not access large-scale compute at all, or that would have to raise dilutive equity capital to fund hardware purchases, may find that sharing a percentage of revenue with Nvidia is cheaper than the alternatives. A startup that could access compute through standard hyperscaler pricing may find that the revenue-sharing arrangement is more expensive over the long term.
The companies Nvidia cited as early beneficiaries of the model — Baseten, Fireworks AI, and Together AI — are inference and model deployment platforms that serve developers, digital natives, and enterprises building AI applications. These are companies whose revenue scales with usage of the underlying compute. For them, a usage-linked payment to Nvidia is structurally aligned with their own revenue model: they pay more when they earn more. That alignment is genuinely useful and may make the model attractive to companies with predictable usage-to-revenue ratios. For companies with more complex revenue structures, or where the path from compute usage to revenue generation is longer and less predictable, the model’s terms matter significantly more.
The Geopolitical and Regional Dimension
Firmus Technologies’ Batam, Indonesia campus — expected to scale to 360 megawatts and 170,000 GPUs — is specifically described as serving AI-native companies that need scalable, energy-efficient compute infrastructure to compete globally. This is the regional AI player use case that the blog post explicitly addresses. Countries and regions that have not had access to hyperscaler-scale AI infrastructure because no hyperscaler has chosen to build data centres there are now potential beneficiaries of the Nvidia-partnered AI factory model. The capital access problem is even more acute for regional players than for US startups, because they also face the challenge of convincing global hyperscalers that their markets are worth the investment.
Nvidia’s willingness to take a revenue share rather than an upfront payment makes the economics of building regional AI infrastructure viable in markets where the revenue trajectory is less certain. It also extends Nvidia’s reach into markets that competing hardware vendors might otherwise be able to address — because the revenue-sharing model only works for Nvidia if the infrastructure is built on Nvidia GPUs, which is the condition of access. The model is simultaneously a financial innovation and a competitive strategy, and the two dimensions are inseparable. For readers tracking how AI infrastructure investment is reshaping the global technology landscape, LiveAIWire’s coverage of SpaceX’s Colossus commercial compute platform and our analysis of AI sovereignty and export controls provides the strategic context in which Nvidia’s model sits.
The Risk That Comes With It
The revenue-sharing model is not without structural risks. Nvidia is extending something analogous to vendor financing — accepting deferred revenue in exchange for a future share of cloud revenue rather than receiving full payment upfront. If the AI cloud companies that build Nvidia-powered factories do not generate the cloud revenue that the model anticipates, Nvidia’s recurring earnings stream from those factories will be lower than projected. The model’s attractiveness to Nvidia is premised on the assumption that AI inference demand will grow substantially and that the factories built under the arrangement will be highly utilised. That assumption is consistent with the consensus view of AI industry growth, but it is an assumption rather than a certainty.
The companies that are building under the model — Sharon AI, Firmus — are committing to large GPU deployments at a moment when the AI infrastructure investment cycle is at a historically high intensity. If that cycle moderates, or if competing hardware vendors reduce the cost of alternative compute significantly, the revenue-sharing model’s terms may need to evolve. Nvidia’s dominance in AI hardware has been the foundation of its extraordinary market valuation. The revenue-sharing model is a bet that dominance can be extended from hardware into the recurring economics of AI services. It is a well-constructed bet. It is not a guaranteed one.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.