NVIDIA teams up with OpenAI to secure 12 gigawatts of computing infrastructure. Huang Renxun: Even if OpenAI does not renew the lease, it can still provide to other customers.

NVIDIA, the leader in AI chips, has revealed its latest developments in AI infrastructure strategy. On August 17, NVIDIA CEO Jensen Huang published a signed article titled Ensuring the Security of Intelligent Infrastructure, stating that NVIDIA has reached a cooperation agreement with energy developer SB Energy to secure land, power, and facility resources located in the PORTS-Pike Technology Park in Portsmouth, Ohio, USA. This site will be dedicated to hosting NVIDIA's computing power systems, while OpenAI will operate as a tenant, constructing and running an AI factory there. In his article, Huang pointed out that AI factories are a decisive infrastructure of the AI era. Currently, the competitive focus of infrastructure has shifted from advanced chips and packaging to land, power, and facilities. The demand for training and inference computing power in AI laboratories is exceedingly remarkable; more computing power means higher intelligence, more products, more users, and more revenue, but its development is severely constrained by inadequate computing power supply. If OpenAI can no longer continue using the park in the future, its computing power can be resold to other qualified tenants within NVIDIA's global ecosystem. Huang emphasized that NVIDIA's computing power is characterized by versatility and interchangeability: Interchangeability drives utilization and durability, making NVIDIA's computing power a rent-able, finance-able productive asset. Therefore, the value of high-quality sites like PORTS-Pike is not limited to a single customer or a single generation of computing power. At the end of the article, Huang outlined NVIDIAs strategic evolution path: from initially manufacturing accelerated computing chips to expanding into systems, networks, and CUDA software, then to a full-stack AI factory, and now further extending to helping clients secure critical underlying resources needed to build these factories. Huang noted that NVIDIA's future layout in LPS will remain strategic and disciplined, with the vast majority of clients continuing to independently acquire LPS resources, and NVIDIA will selectively lock in those excellent sites that have clear, enduring needs and can support multiple generations of computing power upgrades. Subsequently, Huang clarified the financing plan, responding to external skepticism about revolving financing. He pointed out that the company may provide financing support of up to 25% of the project scale for related collaborations: Our role is to help unlock the vast independent capital pools while maintaining a disciplined risk exposure.