NVIDIA Revenue Sharing AI Infrastructure Financing 2026

What NVIDIA's new financing model actually changes

NVIDIA announced the initiative on 1 July in a blog post co-authored by CFO Colette Kress and Raj Mirpuri, positioning it as a response to the industry's shift from training large language models to running continuous, production-scale inference workloads. Rather than simply selling GPUs to cloud providers, NVIDIA is now pairing infrastructure sales with credit support and taking a share of the resulting cloud revenue, creating a recurring income stream tied directly to how much of its hardware is actually consumed.

This matters because emerging AI companies have consistently struggled to raise enough capital to build GPU capacity at the scale production inference demands. By underwriting some of that risk itself, NVIDIA is effectively becoming a financing partner as well as a supplier, which changes the calculus for any buyer trying to work out whether to build, lease, or subscribe to accelerated compute.

Why mid-market UK buyers should pay attention

In our assessment, UK organisations below hyperscaler scale have typically faced a stark choice: commit to years of capital expenditure on GPU clusters, or accept limited, expensive access to third-party cloud capacity. NVIDIA's model, in theory, creates a middle path by making it commercially viable for AI cloud providers to stand up capacity faster and pass access on to smaller customers, including AI-native startups, enterprise IT teams and independent software vendors. We should note this UK-specific angle is our own analysis rather than a point NVIDIA's announcement makes.

For finance and infrastructure teams weighing GPU investment against operating cost, this could plausibly be a genuine planning input rather than just a supplier announcement. Running the numbers through your own organisation's finance and capacity planning tools is a sensible first step before assuming a capital-heavy build is the only option.

Inside the first DSX AI factory buildouts

NVIDIA named Sharon AI and Firmus among the earliest participants in the scheme. Sharon AI plans to deploy up to 40,000 NVIDIA Grace Blackwell GB300 GPUs as part of its expansion, while Firmus is developing a DSX AI factory campus in Batam, Indonesia, with plans to scale to 360MW and support as many as 170,000 NVIDIA GPUs.

These are large, capital-intensive projects that illustrate the scale NVIDIA is targeting with the credit-support model. In our view, UK buyers evaluating whether to build their own clusters or tap into provider capacity should benchmark any in-house plans against comparable reference architectures and against the practicalities covered in server configuration guidance.

The AI-native platform layer driving demand

Alongside the AI cloud providers building DSX factories, NVIDIA highlighted platforms including Baseten, Fireworks AI, and Together AI as examples of AI-native clouds generating demand for on-demand accelerated computing. These platforms support workloads spanning model training, fine-tuning, post-training optimisation, and large-scale inference for enterprise and developer customers.

For UK teams building or buying AI applications, this layer is arguably more relevant day-to-day than the underlying hardware financing, though again this is our own read rather than a point sourced to NVIDIA. It shows there is now a maturing ecosystem of providers competing to deliver inference capacity on flexible terms, which should, in theory, put downward pressure on access costs over time as more capacity comes online through NVIDIA-backed buildouts — though, as flagged below, revenue-share costs could equally be passed through to customers rather than reduce prices.

Diligence points before committing to a provider

Revenue-sharing arrangements are attractive because they lower the barrier to accessing GPU capacity, but they also mean the providers UK buyers contract with are themselves carrying financing and revenue-share obligations back to NVIDIA. That creates a layer of commercial dependency worth scrutinising, particularly around capacity guarantees, pricing stability, and what happens if a provider's own financing terms change.

UK buyers have recently had to think hard about vendor lock-in and exit costs in other parts of the stack, as anyone navigating the VMware alternatives landscape after Broadcom's licensing changes will recognise. The same discipline applies here: understand the contract structure, ask what recourse exists if a provider's capacity or pricing shifts, and avoid assuming access will remain on today's terms indefinitely.

• •Confirm whether capacity commitments are fixed-term or usage-scalable
• •Check how revenue-share costs are passed through in provider pricing
• •Ask about portability if you later need to move workloads elsewhere
• •Map current and near-term GPU demand before signing multi-year terms

What UK infrastructure buyers should do next

In our assessment, this financing model is unlikely to change the fundamentals of GPU scarcity overnight, but it does widen the set of realistic funding pathways for organisations that previously assumed large-scale AI infrastructure was out of reach without hyperscaler-level capital. The practical move now is to size actual workload requirements, compare the cost of provider-backed capacity against direct hardware investment, and factor in components choice via our GPU accelerators resource where in-house builds remain the preferred route.

Given the pace of change in AI infrastructure financing, it's worth getting an outside view before committing capital or signing multi-year capacity agreements. Our engineers are available to talk through your GPU cluster financing options against your specific workload and budget profile.