Just Announced: OpenAI to Build 5GW Data Center for Interstellar Gate, Elon Musk Unveils 5-Year AI Infrastructure Plan}
OpenAI and partners are developing a massive 5GW data center for the Interstellar Gate project, while Elon Musk announces a bold 5-year plan to scale AI infrastructure to 50 million H100-equivalent GPUs.
Yesterday, The Wall Street Journal reported that OpenAI and SoftBank have had disputes over the delayed six-month-old Stargate project, leading to significant reductions in recent plans.
Sources reveal that SoftBank and OpenAI co-led the Stargate project, but disagreements over key terms, including data center locations, have persisted. Despite a January pledge to invest $100 billion immediately, the project now aims to build a smaller data center by the end of this year, likely in Ohio.
In response, OpenAI announced new progress: they are collaborating with Oracle to develop an additional 4.5 GW Stargate data center capacity in the US.
Furthermore, the Stargate I data center in Abilene, Texas, is about to go online. Combined, the total capacity of OpenAI’s Stargate data centers developed with Oracle will exceed 5 GW.
To put this into perspective, according to the 2024 Shanghai Power Supply Sustainability Report, Shanghai’s peak summer electricity load reached 40.3 GW in 2024.
OpenAI states that this 5 GW data center will operate over 2 million chips. CEO Sam Altman previously announced plans to deploy 1 million GPUs by the end of this year (not all in this data center).
OpenAI also shared that the new data center capacity will support job creation, economic growth, and AI development, marking a significant milestone aligned with their long-term vision of making AI benefit everyone. Altman posted photos of the data centers on Twitter.
This is the Stargate I data center currently under construction in Abilene, Texas, with some facilities already operational. Oracle began delivering the first NVIDIA GB200 racks last month, and OpenAI has started early training and inference workloads, aiming to push the boundaries of next-generation AI research.
Regarding the dispute with SoftBank, OpenAI responded: "Besides this expansion with Oracle, our collaboration with SoftBank is also progressing strongly. Both partnerships are crucial to meet our growing computational needs. We are rapidly advancing site evaluations and reimagining data center designs to support cutting-edge AI."
The Stargate project has been ongoing for six months. As OpenAI’s primary AI infrastructure platform, it has established partnerships with Oracle, SoftBank, and CoreWeave, and is attracting international investments through OpenAI for Countries. Microsoft continues to provide cloud services, including via the Stargate project.
OpenAI describes Stargate as an ambitious plan to seize a historic opportunity. With strong support from global partners, governments, and investors, this vision is gradually becoming reality—highlighted by the US White House’s recognition of AI infrastructure’s key role in innovation, economic growth, and competitiveness.
Elon Musk Responds: Our 5-Year Plan
Shortly after OpenAI announced its 5 GW project with Oracle, Elon Musk shared a video from October last year, revealing that xAI’s Colossus 1 supercluster has 230,000 GPUs (including 30,000 GB200 units) for training Grok, with inference handled by cloud providers. Colossus 2 will be launched in a few weeks, with an initial deployment of 550,000 GB200 and GB300 units for training.
Looking further ahead, Musk aims to reach an AI compute capacity equivalent to 50 million H100 GPUs within five years, with a plan to deploy 550,000 GB200 units in Colossus 2, which is 55 times the size of the original Colossus 1.
Some netizens estimated that this would be comparable to 5.5 million H100 GPUs, emphasizing the scale of Musk’s ambitious plan. The images below illustrate the scale and progress of these projects.
It appears that NVIDIA’s stock will continue to rise as these plans unfold. What are your thoughts on these massive-scale projects by OpenAI and Musk?
References: