NVIDIA Invests $1.5 Billion in SB Energy to Secure 8GW of AI Compute Capacity, with OpenAI as Sole Tenant

NVIDIA is pushing its competitive boundaries beyond chips and servers, advancing further into the most foundational layers of AI data centers: land, power, and construction.

On August 17, NVIDIA announced a strategic partnership with SB Energy to provide credit guarantees for the PORTS-Pike technology park in Ohio, transforming it into a hyperscale data center dedicated to NVIDIA's AI computing infrastructure, with OpenAI as the sole tenant. Concurrently, NVIDIA invested $1.5 billion in SB Energy.

By directly entering the underlying infrastructure segment of Land, Power, and Shell (LPS), NVIDIA's move carries significance far beyond typical data center investments. The core shift lies in NVIDIA using its own creditworthiness to "endorse" AI infrastructure financing, thereby securing long-term and exclusive space for compute deployment. This marks the chip giant's expansion from a pure AI hardware supplier to a core participant and organizer of "AI factory" infrastructure.

The project's initial deployment capacity is 4.25 IT-GW, with NVIDIA holding an option to expand by an additional 3.75 IT-GW, bringing the total planned capacity of the park to 8 IT-GW. NVIDIA disclosed that each generation of AI factory systems deployed at PORTS-Pike will include approximately 1.5 million GPUs, corresponding to roughly $150 billion to $200 billion in revenue opportunities for NVIDIA.

From Selling Chips to Locking In "AI Factories"

NVIDIA refers to AI factories as "the decisive infrastructure of the intelligent era" and views LPS as the next category of strategic resources following chips, packaging, memory, and networking. The logic mirrors traditional supply chain management: locking in key input factors in advance when customer demand visibility is high.

PORTS-Pike will exclusively deploy NVIDIA's full-stack DSX AI factory platform, covering GPUs, CPUs, networking, and infrastructure software. For NVIDIA, once the park is secured, long-term assets such as underlying land, power, and buildings can support continuous upgrades across multiple GPU generations, rather than losing value as a single chip generation's lifecycle ends.

In other words, NVIDIA is attempting to transform AI infrastructure from a one-time "build server rooms, buy GPUs" expense into a long-term asset that can continuously roll forward with GPU generational updates.

OpenAI Demand Becomes the Project's Pivot Point

Behind this collaboration is the reality that the compute demands of frontier AI companies are rapidly exceeding the carrying capacity of traditional balance sheets.

NVIDIA explicitly pointed out that the growth rate of compute demand for some frontier AI labs has surpassed the scope supported by their own balance sheets and long-term credit profiles, making it difficult for them to independently finance large-scale infrastructure.

The PORTS-Pike agreement effectively provides a new financing model: NVIDIA uses its scale, creditworthiness, and long-term demand visibility to provide partial credit support for infrastructure financing, while OpenAI locks in compute capacity through long-term leases.

In terms of compute scale, OpenAI has currently committed to deploying approximately 12 gigawatts (GW) of NVIDIA compute capacity by 2030. According to NVIDIA's disclosures, if the PORTS-Pike arrangement expands beyond the initial 4.25 GW, OpenAI's related deployment scale is expected to rise to approximately 16 GW.

NVIDIA estimates that at this scale, the associated compute opportunities will correspond to about $600 billion in NVIDIA's computing business revenue by 2030. In other words, PORTS-Pike is not merely an infrastructure financing arrangement; it is NVIDIA further binding compute procurement, infrastructure construction, and financing capabilities around core customer needs.

This also explains why NVIDIA is willing to intervene in infrastructure financing, which was previously not its core business: as long as customer demand is sufficiently certain, locking in LPS equates to securing deployment space for future generations of GPUs in advance.

How the 8GW Park Will Be Implemented

PORTS-Pike is located in Pike County, Ohio, adjacent to the site of the decommissioned Portsmouth Gaseous Diffusion Plant. The park spans both private and federal land, with development led by SB Energy and partners including AEP Ohio, the U.S. Department of Energy, and the U.S. Department of Commerce.

The project will be constructed in phases, with the first phase expected to come online sequentially between 2028 and 2030. SB Energy will provide data center construction, operation, and hosting services to OpenAI through a 20-year lease.

From a business model perspective, OpenAI is responsible for long-term leasing and usage, SB Energy handles park development and operations, while NVIDIA ensures the continuous deployment of NVIDIA compute capacity in the park through investment and limited credit guarantees.

The project also includes significant community investment, including an initial $80 million community benefits fund and hundreds of millions in long-term community investment commitments, and is expected to create tens of thousands of jobs.

NVIDIA's Guarantee Is Not a "Bailout"

What the market needs to focus on more is the extent of risk NVIDIA has actually assumed.

NVIDIA explicitly stated that its credit support does not cover the entire cost of the park, nor does it bail out all of OpenAI's lease obligations. Instead, it is limited to specific portions of lease and power payments, as well as a specific residual value commitment.

As OpenAI pays rent and the data center gradually comes online, NVIDIA's contingent exposure will decrease accordingly.

If OpenAI ceases to use part of the capacity in the future, NVIDIA believes that based on the broad adaptability of the CUDA ecosystem, these compute facilities can still be subleased to cloud service providers, enterprise customers, other AI labs, and startups. Therefore, the park itself does not rely entirely on the long-term creditworthiness of a single tenant.

This is also why NVIDIA emphasizes the "strategic and disciplined" nature of its LPS strategy: the company is not prepared to fully bear the data center construction costs for customers, but rather only supports core sites where demand is highly visible, lasts long enough, and can support the deployment of multiple generations of NVIDIA compute.

AI Infrastructure Competition Enters the Next Phase

In recent years, the core of competition in the AI industry has been GPUs, HBM, advanced packaging, and high-speed networking; as compute demand continues to explode, land, power, and data center construction cycles are becoming new bottlenecks.

Therefore, the significance of PORTS-Pike lies not just in being an 8GW data center project.

More importantly, NVIDIA is attempting to establish a complete "chip + networking + software + infrastructure" AI factory system. For customers, this means shifting from purchasing GPUs to directly obtaining a complete set of AI compute infrastructure capable of sustainable upgrades; for NVIDIA, it means further binding future GPU demand to its own platform by locking in LPS resources in advance.

If this model can be replicated, NVIDIA's role may undergo a significant change: it will no longer just sell GPUs to data centers, but will begin to participate in deciding "where these GPUs will be deployed, who will build them, and what power source will run them."

This may well be the most noteworthy aspect of the PORTS-Pike project.