OpenAI has built five computing power centers in one go, and NVIDIA has satisfied Masayoshi Son and Oracle.

NVIDIA has just planned a new investment of $10 million in OpenAI, and OpenAI has announced how it will use the money: it will cooperate with Oracle and SoftBank to build five data centers at once.

These newly built data centers will be part of Altman's "Stargate" project, and the planned capacity of the project will be increased to nearly 7GW, roughly equivalent to seven large nuclear reactors.

OpenAI said that the new data centers will enable the "Stargate" project to be completed ahead of schedule.

OpenAI purchases Oracle's cloud services → Oracle buys NVIDIA GPUs → NVIDIA invests in OpenAI. With the announcement of the new plan, this triangular relationship has been strengthened once again.

In addition, Altman also revealed in his personal blog that OpenAI will have new plans, partners, and some "new ideas" in computing power construction in the next few months.

But have you noticed in this cooperation that Microsoft, the former close ally, has little to do now.

Microsoft stumbles, Oracle feasts.

Co - building Data Centers with Oracle

The five new data centers officially announced by OpenAI will be jointly developed by OpenAI, Oracle, and SoftBank.

Three of them will be built in cooperation with Oracle, located in Shackelford County, Texas, Doña Ana County, New Mexico, and an undisclosed location in the Midwest respectively.

The other two will be operated by SB Energy, a subsidiary of SoftBank that develops solar and battery projects. They are located in Lordstown, Ohio, and Milam County, Texas respectively.

Of course, OpenAI is already familiar with cooperating with Oracle. The data center in Abilene, Texas, was jointly built by the two parties, and in July, they reached an agreement to expand this data center, aiming to develop a data center capacity of up to 4.5GW.

The operation model is "lease - to - own" - the data center in Abilene is mainly owned and operated by Oracle Corporation, and OpenAI uses it as the main tenant.

People familiar with the matter revealed that the data center is currently running on Oracle Cloud Infrastructure and supports OpenAI's training and inference workloads.

Oracle is currently building eight data center halls in Abilene, each about 100MW. After completion, the facility will accommodate more than 400,000 GPUs, with a scale of 1.4GW.

NVIDIA's $100 Billion Investment in OpenAI

As another corner of the "triangular relationship", NVIDIA announced yesterday that it plans to invest $100 billion in OpenAI to build a 10GW data center, which is roughly equivalent to 4 - 5 million GPUs.

However, this $100 billion investment will not be made all at once, but in batches. NVIDIA will invest $10 billion each time OpenAI completes 1GW of facilities. The first batch is planned to be completed in the second half of next year.

$100 billion, one - fifth of OpenAI's valuation, is indeed a significant amount, but is it really enough for OpenAI?

During the earnings conference call in August, Jensen Huang said that the cost of building an AI data center with a capacity of each GW is about $50 billion.

But NVIDIA's investment averages only $10 billion per GW, and it will be given after the facilities are completed. Where to get the remaining money has become a major question for people.

OpenAI did not respond to requests for comments on its financing plan, and neither NVIDIA nor OpenAI has given a specific timeline for the full 10GW capacity to go online.

Before NVIDIA announced its investment, the debt rating agency Moody's also had similar doubts, suspecting whether OpenAI had enough cash flow to pay for Oracle's contracts.

However, NVIDIA's investment news has given a boost to OpenAI and increased the certainty that OpenAI will purchase cloud resources from Oracle as planned.

One More Thing

OpenAI initially positioned Stargate as a "new company", and SoftBank CEO Masayoshi Son would serve as its chairman.

However, according to an executive close to the project, Stargate is actually a general term referring to all of OpenAI's data center projects.

But this executive specifically pointed out that it does not include projects developed in cooperation with Microsoft.

Microsoft, an old - time investor in OpenAI, seems to have been excluded from the blueprint of the Stargate project.

The honeymoon period between Microsoft and OpenAI is over.

Reference Links:

[1]https://openai.com/index/openai-nvidia-systems-partnership/

[2]https://www.wired.com/story/openai-oracle-softbank-data-center-stargate-us/

[3]https://blog.samaltman.com/abundant-intelligence

[4]https://www.reuters.com/business/more-questions-than-answers-nvidias-100-billion-openai-deal-2025-09-23/

This article is from the WeChat official account "QbitAI", author: Cressey. It is published by 36Kr with permission.