#NvidiaAndOpenAISecure12GWCompute

NVIDIA + OpenAI: The AI Race Is Becoming a Power Race

The most important number in the latest NVIDIA–OpenAI story may not be the number of GPUs.

It is 12GW.

NVIDIA CEO Jensen Huang said OpenAI’s existing and planned commitments for NVIDIA computing capacity are now approximately 12GW, while the broader PORTS-Pike opportunity in Ohio could potentially approach 16GW if the project expands further.

That changes how we should think about the AI infrastructure race.

A few years ago, AI infrastructure discussions were dominated by questions like: How many H100s? How many Blackwell GPUs?

Now the conversation is increasingly about something much bigger:

How much power can you secure, where can you build, and how quickly can you turn that electricity into usable computing capacity?

This is why the PORTS-Pike project is particularly interesting.

SB Energy is expected to build, own and operate the data-center infrastructure, while OpenAI has committed to a long-term lease. NVIDIA is also providing financial support for parts of the land, power and infrastructure development and has announced a $1.5 billion investment in SB Energy.

That means NVIDIA’s role is evolving.

It is no longer simply a company selling chips into data centers. It is increasingly becoming part of the broader infrastructure ecosystem required to deploy those chips at enormous scale.

And that is where the concept of the AI Factory becomes important.

The GPU is only one component. You also need land, electricity, substations, cooling, networking, buildings, financing and long-term infrastructure planning.

In other words, the bottleneck is increasingly shifting from:

“Can we get enough GPUs?”

to:

“Can we build enough power-backed computing capacity?”

There is another important distinction.

The 12GW figure represents OpenAI’s NVIDIA computing commitments; it should not be interpreted as the capacity of one Ohio facility or as OpenAI’s entire computing footprint.

OpenAI is also pursuing additional infrastructure through other technology, cloud and chip partners. That diversification matters because frontier AI companies cannot afford to depend on a single source of computing capacity as demand continues to expand.

But the model also creates risks.

Long-term infrastructure commitments require enormous capital, and assumptions about future AI demand are being embedded into projects that may operate for decades. If AI demand continues accelerating, these facilities could become strategic assets. If demand expectations change dramatically, some of those long-term commitments could face significant repricing.

That is why the NVIDIA–OpenAI relationship is becoming more interesting than a simple chip-supply agreement.

The AI race is increasingly becoming a competition for chips + power + land + capital + infrastructure.

And when the industry starts measuring its ambitions in gigawatts rather than individual GPUs, it is clear that AI is entering a completely different phase of expansion.

The next major AI bottleneck may not be computing technology.

It may be electricity.

$NVDA