OpenAI Announces 5 New Stargate Hubs in the U.S. to Support Knowledge Infrastructure

Since Microsoft and OpenAI announced on March 30, 2024, that they were planning a $100 billion supercomputer campus called Stargate, I have written several articles about this initiative. Since then, construction of overseas hubs has begun in Northern Europe and India, and now that five new Stargate hubs have finally been announced within the United States, I will dive deeper into them.

What is Stargate?

OpenAI has long been promoting a large-scale infrastructure initiative to support AI under the name "Stargate."

It is a project to update the "map of computing power" that supports AI training and inference by deploying GPU clusters and data center groups on a national scale. In this article, I will discuss the announcement made yesterday that OpenAI, Oracle, and SoftBank will collaborate to establish five new Stargate hubs in the United States.

This move is not merely an "expansion," but clearly reveals an intention to push the front lines of the AI "infrastructure race" within the U.S. It can also be read as a piece of the answer to the question of how to geographically and institutionally distribute computing resources in anticipation of the coming AGI era.

Overview and Significance of the New Hubs

New Hubs and Numerical Targets

The five hubs announced this time have the following characteristics:

• Location: Various places across the U.S., such as Texas, New Mexico, and Ohio.

• Scale: Over 5.5 gigawatts (GW), with cumulative investment aimed at approaching the full target of 10GW in the future.

• Funding Goal: A strategy to achieve the massive investment target of $500 billion ($500B) ahead of schedule by the end of 2025.

• Employment and Regional Contribution: Aiming for over 25,000 local jobs at the time of hub establishment, plus ripple effects in the surrounding areas.

Looking at these figures, I believe this is not just an announcement that "we are increasing data centers for AI," but a move involving a "tectonic shift" in the national-scale infrastructure race.

Collaboration Structure: OpenAI × Oracle × SoftBank

To operate this five-hub initiative, cloud infrastructure that can provide the massive computing power required by AI models with "reliability, scalability, and cost-efficiency" is essential. Oracle is based on its cloud business (OCI), and it is positioning that as the pillar of Stargate.

SoftBank (including SB Energy) plays the role of securing technology and resources in terms of power supply, energy design, and site construction and design. In other words, this means that the vision includes not only AI infrastructure—GPUs and chips—but also a comprehensive design that encompasses "stable power supply" and "construction, cooling, and operational efficiency."

Why rush the expansion now? Decoding the strategic motives

Competitive Pressure and the Value of the "First Move"

AI researchers, startups, and cloud providers are already engaged in a "battle for computing power." If they cannot secure the resources to run large-scale models, they will fall behind. In that sense, the Stargate expansion can be read as a move to practice the iron rule of "first-mover advantage" at the infrastructure level.

Cost Optimization and Efficiency

By distributing sites extensively, it becomes possible to implement strategies that optimize cooling efficiency, power procurement, and leverage regional electricity rate fluctuations and climate conditions. For example, utilizing cold climates to lower cooling costs or taking advantage of contract terms with local utility companies.

Resilience and Scalability

Having sites across a wide area makes it easier to mitigate the risk of system-wide outages in the event of power failures, natural disasters, or policy changes in a single region. It also provides the flexibility (scalability) to increase or decrease capacity in immediate response to demand.

Potential Risks and Challenges Cannot Be Ignored

• Massive Capital Investment: A sum of $500 billion carries the risk of stranded capital if the project does not succeed.

• Legal and Regulatory Barriers: Projects are susceptible to the influence of land, power, and local government coordination within the U.S., as well as environmental regulations, subsidies, and tax systems.

• Sustainability of Energy Sources: GPU clusters consume vast amounts of electricity. The question is how to ensure the use of renewable energy or the decarbonization of regional power.

• Technical and Physical Limitations: If fine-grained infrastructure preparation—such as cooling design, network latency, operational costs, and equipment failure tolerance—is neglected, it will become a bottleneck.

• The Trade-off Between Centralization and Decentralization: Excessive decentralization increases communication latency and synchronization overhead. Design decisions must be made to strike a balance between 'distance/latency × throughput' somewhere.

Future Predictions and Questions Raised by This Announcement

• OpenAI is turning the strategy that 'those who control the infrastructure will be the ones who run AI' into reality.

• In the future, we can assume a scenario where this Stargate network is expanded to an international level (for example, articles about a UK version of Stargate have already been found).

• However, when AGI emerges, it may become a strategic resource akin to 'infrastructure as military power.' The distribution of data centers will likely become inseparable from national defense and security.

• Given this scale of investment, the design and research roadmap for AI models may be constrained by 'how much can be trained.' We are in an era where computational resources dictate research strategy.

Still 'Not Enough'

Prior to this announcement, Sam Altman stated, 'A $100 billion scale is still small. It only accounts for a few million GPUs, and what is truly needed is on the scale of billions. We are three orders of magnitude away from where we need to be.'

From this perspective, while the five new Stargate sites being built in the U.S. are certainly a major step forward, they can still be considered 'in progress.' Even if the five sites cumulatively achieve a scale approaching 10GW, that is merely the 'minimum infrastructure to welcome the dawn of AGI.'

Considering an era where AI enters the core of social systems—such as education, medicine, research, and energy design—on a human scale, further expansion by several or dozens of times will be inevitable. The overseas sites started in Northern Europe and India, along with these five U.S. sites, are just 'stepping stones to reach the next order of magnitude.'

In other words, even when the five Stargates are completed, it will not be the goal, but rather the stage where we can finally say we have 'lined up at the starting line.' The infrastructure race to support the AGI era has only just begun.

This is just an image, but I made a Stargate MV before, so please take a look if you have time.