NVIDIA’s New Deal: GPUs Now, Revenue Later
NVIDIA has introduced a revenue-sharing and credit-support model that lets AI cloud providers deploy large-scale GPU infrastructure without full upfront payment — collecting a share of cloud revenue in return. The first two partners, Sharon AI and Firmus Technologies, are together committing up to 210,000 Grace Blackwell GB300 GPUs under the programme, announced July 1, 2026. The model is part of NVIDIA’s broader DSX AI factory strategy targeting AI-native startups, model builders, and inference providers that have historically struggled to access capital-intensive compute at scale. Read how export control changes are reshaping the wider AI infrastructure landscape.
How the DSX Revenue-Share Model Works
Three actors, one loop — NVIDIA earns on hardware and again on cloud usage.
Emerging AI companies have historically faced a structural barrier: even long-term compute commitments were rarely sufficient to unlock the financing needed for large-scale GPU infrastructure. NVIDIA’s new model addresses this directly. Under the arrangement described in the NVIDIA blog post co-authored by CFO Colette Kress, AI cloud providers procure NVIDIA infrastructure for AI-native, enterprise, and ISV customers through an economic alignment structure that includes both revenue-sharing and credit support. This gives cloud operators a capital-efficient path to scale while giving NVIDIA a recurring, usage-linked earnings stream alongside standard product revenue.
The shift matters because AI compute is moving from model development into continuous production inference — what NVIDIA describes as “AI factories” generating tokens at scale around the clock. That type of workload demands always-on, multi-tenant accelerated computing that can come online quickly and stay highly utilised. For context on how AI inference demands are expanding across enterprise and government, the infrastructure buildout required is substantial.
First Partners Under the Programme
Click each tab to explore the details of each deployment.
“This strategic collaboration with NVIDIA marks a pivotal moment in Sharon AI’s mission to deliver sovereign, large-scale AI compute infrastructure.”
— James Manning, Co-Founder and CEO, Sharon AI“AI-native companies need access to scalable, energy- and cost-efficient compute infrastructure to compete globally. Firmus AI cloud is building a NVIDIA DSX-aligned AI factory, which will enable our cloud to help more customers access the compute they need to build and scale AI.”
— Tim Rosenfield, Co-CEO, Firmus TechnologiesWhere the Infrastructure Is Being Built
Click the markers for deployment details.
Who This Is Built For
NVIDIA named these AI-native companies as the type of customers the programme targets.
The DSX model pairs NVIDIA’s full-stack AI factory platform with Firmus’s proprietary HyperCube liquid-cooled architecture. NVIDIA says the GB200 NVL72 rack design increases compute density and reduces floor space requirements compared to air-cooled infrastructure, with GB200 delivering up to 25 times more performance at the same power compared to H100 air-cooled systems. Construction costs for AI-optimised data centres with advanced liquid cooling have reached approximately $20 million per megawatt in industry estimates, making the credit-support component of NVIDIA’s model material for operators who cannot fund builds outright.
Separately, NVIDIA has made over $40 billion in direct AI equity investments so far in 2026, according to CNBC and TechCrunch, including a roughly $30 billion investment in OpenAI. The revenue-sharing compute model operates on a different track — it creates recurring income tied to platform usage rather than a balance sheet equity stake. On the agentic AI side, see how the infrastructure being built here intersects with the security risks that come with large-scale agentic deployments.
NVIDIA’s Two Revenue Streams Under This Model
For every dollar of cloud usage on supported capacity, NVIDIA earns twice.
Bar widths are illustrative, not based on disclosed financial proportions. Revenue share percentage undisclosed by NVIDIA.
What DSX Actually Is
Breaking down NVIDIA’s full-stack AI factory platform.
Coverage Summary
NVIDIA’s revenue-sharing and credit-support programme was announced on July 1, 2026, through a blog post co-authored by CFO Colette Kress and Raj Mirpuri. The programme was covered in the context of Sharon AI and Firmus Technologies as its first two participating cloud operators, with a combined commitment of up to 210,000 Grace Blackwell GB300 GPUs across deployments in Australia and Batam, Indonesia.
The Firmus Batam campus was covered separately from the July 1 announcement — that partnership, running through 2034, was announced earlier and encompasses up to 170,000 NVIDIA accelerators across Grace-Blackwell, Vera-Rubin, and Vera platforms, with projected customer offtake of $25 to $30 billion over the first six years based on Firmus’s own customer commitment data. The facility is co-developed with DayOne and is targeted to be operational in Q1 2027.
The structural details of the model — dual revenue streams, credit support for operators, and usage-linked earnings — were reported as described in NVIDIA’s official blog and related primary disclosures from Sharon AI (SEC Form 8-K) and Firmus Technologies. The revenue share percentage was not disclosed by NVIDIA. For further context on AI infrastructure investment patterns, see how major technology companies are repositioning capital across infrastructure categories.
