On July 6, NVIDIA officially announced the launch of a new Revenue-Sharing Model, allowing AI startups to exchange future revenue for GPU computing power without bearing the upfront huge hardware procurement costs; Sharon AI and Firmus are the first partners.
First Partners: Sharon AI Deploys Up to 40,000 GB300 GPUs
According to the official NVIDIA announcement, the first partners to respond to this business model are: Sharon AI, which is deploying up to 40,000 NVIDIA Grace Blackwell GB300 GPUs; Firmus, which is building a DSX AI factory campus in Batam, Indonesia, expected to expand to 360 MW power capacity to accommodate up to 170,000 NVIDIA GPUs.
Sharon AI co-founder and CEO James Manning said: "The strategic partnership with NVIDIA is a key moment for Sharon AI to realize its vision of sovereign-scale AI computing power." Firmus Technologies co-CEO Tim Rosenfield noted: "AI-native companies need scalable, energy- and cost-efficient computing infrastructure to compete globally."
Revenue-Sharing Model Three-Tier Structure: Standard Hardware Revenue, Operating Profit Sharing, and No Prepayment for Computing Power
According to NVIDIA's official explanation, the newly launched Revenue-Sharing Model consists of three tiers:
First tier, NVIDIA's cloud partners provide GPU computing services to AI startups;
Second tier, NVIDIA takes a percentage of its partners' cloud operating revenue, not just relying on one-time hardware sales;
Third tier, AI startups offset computing costs with future revenue, without needing to purchase GPU hardware upfront.
The core difference from the traditional model is: in the traditional model, AI startups have to purchase hardware themselves or sign long-term cloud leases, and with the rising demand for Blackwell-architecture GB300 GPUs, the financial pressure is extremely heavy.
AI-Native Startups Like Baseten, Fireworks AI Are Potential Beneficiaries
According to NVIDIA's official blog and reports, the potential beneficiaries and target groups of this model are as follows:
AI-native startups (such as Baseten, Fireworks AI, Together AI): Need immediate access to AI cloud computing for model training, post-training fine-tuning, and high-concurrency inference, but are in the transition phase from pilot to production, with business models not yet fully formed.
Regional AI operators: Mainly focused on the long-tail market underserved by large cloud providers such as AWS, Azure, and GCP.
Enterprises and research institutions: Need sustained computing support but cannot afford upfront hardware investment.
Sovereign AI participants: Sovereign computing capacity building represented by regional AI bases such as Firmus in Batam, Indonesia.
July 2 Batam Pilot Upgraded to Global Strategy, NVIDIA Transitions from Chip Seller to AI Ecosystem Operator
According to reports, the cooperation pilot between NVIDIA and Firmus in Batam, Indonesia was first revealed on July 2, described at the time as a "single-site experiment"; on July 6, NVIDIA officially upgraded it to a global institutionalized business strategy, turning from a case into a standardized product. Through the combination of "revenue sharing + credit support", NVIDIA not only creates a more stable recurring revenue stream, but also extends its ecosystem from top-tier large cloud providers to the bottom long-tail market, marking NVIDIA's formal transformation from a chip seller to an AI infrastructure ecosystem operator.
Frequently Asked Questions
How does NVIDIA's Revenue-Sharing Model work?
According to NVIDIA's official explanation, NVIDIA's cloud partners provide GPU computing services to AI startups; AI startups offset computing costs with future revenue, without upfront hardware purchase; NVIDIA takes a percentage of its partners' cloud operating revenue, with its interests directly tied to the ecosystem's long-term operational performance.
Who are the first partners to join the Revenue-Sharing Model?
According to the official NVIDIA announcement, the first partners are Sharon AI (deploying up to 40,000 NVIDIA Grace Blackwell GB300 GPUs) and Firmus (building a DSX AI factory campus in Batam, Indonesia, expected to expand to 360 MW power capacity, accommodating up to 170,000 NVIDIA GPUs).
What type of AI companies are the main target audience for this model?
According to NVIDIA's official blog, the main targets are AI-native startups (such as Baseten, Fireworks AI, Together AI), regional AI operators, and companies in the transition phase from pilot to production that need immediate access to computing power but cannot afford upfront hardware investment.
