NVIDIA invites capital partners to build AI infrastructure at scale

What happened?

NVIDIA has announced a new business model to accelerate the construction of AI infrastructure at scale, called DSX AI factories. The company invites capital partners (AI cloud companies) to build and operate these 'AI factories' that generate tokens continuously, offering in return a revenue-sharing model and credit support. This approach aligns the economic interests of NVIDIA and its partners, enabling faster deployment of computing capacity.

The first partners to join are Sharon AI and Firmus. Sharon AI will deploy up to 40,000 NVIDIA Grace Blackwell GB300 GPUs, while Firmus will build a campus in Batam, Indonesia, scaling to 360 MW and up to 170,000 NVIDIA GPUs. These centers will be designed to serve startups, enterprises, researchers, and regional AI players, providing access to cutting-edge infrastructure.

Why is it important?

Historically, AI startups had limited access to capital-intensive infrastructure. Even long-term commitments were not enough to unlock financing for compute, according to NVIDIA's blog. This new model removes that barrier, allowing more players to access large-scale accelerated compute without waiting for site selection, power acquisition, construction, and hardware deployment.

Moreover, the shift of AI from model development to production inference is increasing demand for continuous compute. AI factories operating 24/7 generating tokens are the new paradigm. NVIDIA positions itself as the key enabler of this transition, offering a business model that accelerates adoption of its platforms among customers, while providing capital partners an efficient route to scale and NVIDIA a recurring revenue stream tied to usage.

This move is comparable to the transition of traditional cloud computing, where companies like AWS, Azure, and Google Cloud built massive data centers to rent capacity. However, here NVIDIA not only provides the hardware but also participates in the financial risk and operating revenue, representing an innovation in the semiconductor industry's business model.

What consequences will it have?

For NVIDIA, this model generates a recurring revenue stream tied to usage, similar to a subscription model, which could smooth the volatility of its hardware sales. For capital partners like Sharon AI and Firmus, it offers an efficient route to scale without assuming all the financial risk, as NVIDIA shares credit and revenue risk. For end customers (model builders, inference providers, agent platforms, enterprises), it means faster access to cutting-edge compute, reducing timelines from months to weeks.

The initiative could also accelerate adoption of NVIDIA platforms among startups and regional companies, expanding its ecosystem. However, it also raises questions about dependence on a single supplier and market concentration. If AI factories become the standard, NVIDIA could further consolidate its dominance in the AI accelerator market, which already exceeds 80% share according to analyst estimates.

Additionally, this model could have geopolitical implications, as seen in the case of Firmus in Indonesia. By building infrastructure in emerging regions, NVIDIA and its partners can help distribute AI compute capacity beyond traditional hubs in the US and China, potentially easing tensions over concentration of critical resources.

What should readers know?

This announcement confirms that NVIDIA is not just selling GPUs but building a complete ecosystem around its hardware. The revenue-sharing model is a financial innovation that could be replicated in other tech sectors, such as storage or networking. For AI startups, this is an opportunity to access infrastructure previously reserved for tech giants like OpenAI or Google.

It is important to follow the evolution of projects like Sharon AI and Firmus, as they will serve as proof of concept. If successful, we could see a wave of AI factory construction driven by this model, with potential to transform the AI economy. However, readers should be cautious: long-term viability will depend on sustained demand for AI tokens and NVIDIA's ability to maintain its technological edge against competitors like AMD, Intel, and custom chip startups.

In summary, NVIDIA's announcement is a milestone combining technological and financial innovation, with the potential to democratize access to AI infrastructure, but also to deepen dependence on a single supplier. Time will tell whether this model becomes the industry standard or a failed experiment.