Nvidia is promising as much as $105 billion to back OpenAI‘s plans to lease a large new data center in Ohio, one of the biggest infrastructure-related pledges from the chipmaker. Developed by SB Energy, which is backed by SoftBank, the facility is anticipated to become key to OpenAI’s expanding computing framework. In addition to the guarantee, Nvidia will also be making a separate $1.5 billion investment in SB Energy, further driving the project.
The agreement is a testament to the close relationship between the evolution of AI and the creation of large-scale physical structures. Advanced AI systems need a massive amount of computing power, which relies on special components, data centers, electricity, land, and connections to the power grid. Nvidia, whose graphics processing chips are used to train and run AI models, is seeing more involvement in the supply of these chips as well as the infrastructure needed to use them at scale.
The Ohio project is to be in Pike County with a potential capacity of as much as 8 gigawatts. The first 800 MW will be in service in 2028. The site will be leased to OpenAI for 20 years, providing the companies with a long-term plan for the development of computing power. Nvidia is also poised to become the sole chip vendor for the site, meaning the company’s investment in its infrastructure will go hand-in-hand with its AI processors’ future demand.
However, Nvidia’s CEO Jensen Huang has reacted to the deal with respect to any criticism that it may be circular financing. As technology firms, chip makers, infrastructure builders and investors continue to pour capital into each other, the problem has gained momentum in efforts to hurriedly build AI facilities. There have been doubts that such deals can form a vicious circle where companies fund projects they are then asked to supply, once the project is finished.
Huang dismissed that description of the Ohio agreement. We are putting in place long-lived infrastructure for Nvidia compute so that OpenAI can deploy the most productive AI factories that can be repeatedly upgraded with each new generation making them more intelligent and economically viable, Huang said. He also claimed that Nvidia was leveraging its “scale and long-term visibility” to help develop infrastructure that will be useful over multiple generations of its technology.
Financial arrangements have not yet been completely determined for the Ohio project. Sources close to the discussions say it is to be expected to have equity and debt. Future SB Energy public offering, and direct investment from SoftBank may be potential equity sources. The equity financing will be determined and the rest of the financing may consist of project finance loans and possibly public debt or bonds.
Nvidia’s promise doesn’t imply that the company will cover the full expense of the facility or OpenAI’s financial obligations. Rather, the guarantee is supposed to pay some lease and power costs and safeguard a minimum worth of the site. The deal would leave OpenAI liable for rent. If OpenAI fails to fulfill its promise, Nvidia may be responsible for the shortfall between the minimum price and the owner of the property could be able to recover by leasing or selling the property.
It’s also a strategy for Nvidia to secure key supply-chain spots for its computing solution. The company has stated it will take a “selective” land grab of prime data center locations where its chips will be able to run over multiple generations of technology enhancements. This approach may grow more crucial as artificial intelligence firms struggle to secure power supplies, land and high-speed computing power.
One of the biggest problems with the AI development industry is highlighted by the Ohio development. Creating a complex data center is no longer just about purchasing servers and putting the computers in. The facilities use vast amounts of electricity and so have access to reliable power as one of their greatest constraints for future expansion. In some areas in the US, aging power grids are already straining and in other areas, communities have voiced concerns about electricity costs, water use, and the environmental impacts of large data center construction.
SoftBank and SB Energy will build at least 10GWs of new power generation to support the Ohio project. They will also invest about $4.2 billion in regional grid infrastructure in a partnership with AEP Ohio. These investments illustrate the scope of investments needed to enable AI computing. A data center can be located in a particular geographic area, while the power generation demands can affect the adjacent power network and the geographical area.
Financial returns seem to be huge for Nvidia as well. The Ohio facility, which was originally projected to be 4.25 gigawatts, could produce up to $200 billion in profit for Nvidia, Huang has claimed. More generally the company believes that its partnership with OpenAI could generate a total of $600 billion in revenue over the next 10 years by selling computing power of 16 gigawatts. Those statistics are a testament to the value Nvidia is placing on its position in infrastructure that enables the development of AI.
The Ohio project will also bring about significant economic benefits to the region. OpenAI announced that the project could generate about 35,000 construction jobs by 2032, and about 2,500 permanent operating jobs if the facilities are operating. SoftBank and OpenAI have also pledged $80 million towards community projects, which are meant to support the development from within.
Support from both the local community and government will help overcome one of the biggest challenges often encountered in large data center projects: community opposition. The Ohio development is on federal land and is receiving assistance from the U.S. Departments of Commerce and Energy. The project has also been welcomed by state officials on the basis of the jobs and investment potential it offers. The project would benefit from strong political and community support, which would help guard against the delays plaguing other big infrastructure projects.
