Nvidia will invest up to $100B in OpenAI to finance data center construction

Shares of Nvidia Corp. rose nearly 4% today after it announced plans to invest up to $100 billion in OpenAI.

The funding is intended to help the artificial intelligence provider grow its data center capacity. According to OpenAI, the plan is to add least 10 gigawatts’ worth of computing infrastructure. One gigawatt corresponds to the energy use of several hundred thousand homes.

Nvidia plans to disburse the funds “progressively as each gigawatt is deployed.” OpenAI expects to complete the initial phase of the construction project in the second half of 2026. It didn’t specify how many gigawatts’ worth of infrastructure will be built during that initial phase, but disclosed the hardware will be powered by Nvidia’s upcoming Vera Rubin chip.

The Vera Rubin combines a graphics processing unit with a central processing unit in a single package. The CPU, Vera, features 88 cores based on an Arm Holdings plc design. It’s linked to the attached GPU by an interconnect that can move 1.8 terabits of data between the two chips every second.

The GPU in the Vera Rubin chip is based on a new graphics card architecture called Rubin. Nvidia plans to launch two different Rubin GPUs next year. Each chip is optimized for a different subset of the computations involved in running inference workloads.

Nvidia previewed one of the upcoming GPUs, the Rubin CPX, earlier this month. It’s optimized for the initial set of computations that an AI model carries out after it receives a user prompt. Those initial computations require less memory than the subsequent processing steps.

According to Nvidia, the Rubin CPX comprises a single die with 128 gigabytes of GDDR7 memory. It includes circuits optimized to run language models’ attention mechanism and video processing workloads. Nvidia hasn’t shared the specifications of the other Rubin GPU  that it plans to launch.

It’s unclear how OpenAI plans to implement Nvidia’s processors in its data centers. One possibility is that it will use the Vera Rubin NVL144 CPX, an upcoming appliance the chipmaker detailed earlier this month. The system includes 288 GPUs and 36 CPUs that can provide 8 exaflops of performance.

Nvidia and OpenAI stated that they will “work together to co-optimize their roadmaps for OpenAI’s model and infrastructure.” That hints the companies may collaborate on certain engineering initiatives. Other market players have made similar moves. After OpenAI rival Anthropic PBC raised $4 billion from Amazon Web Services Inc. last November, it announced plans to contribute to the development of the cloud giant’s Neutron AI model compiler.

“NVIDIA and OpenAI have pushed each other for a decade, from the first DGX supercomputer to the breakthrough of ChatGPT,” said Nvidia Chief Executive Officer Jensen Huang. “This investment and infrastructure partnership mark the next leap forward—deploying 10 gigawatts to power the next era of intelligence.”

Nvidia and OpenAI have outlined the terms of the partnership in a letter of intent. They plan to finalize the agreement in the coming weeks.

Photo: Nvidia