Earlier, the media reported that xAI had rented Nvidia chips from Oracle for xAI's supercomputing plan for many years. On Tuesday, the media reported that the negotiations between xAI and Oracle had reached a deadlock because Oracle was concerned that the supercomputer site could not get enough electrical utilities supply, and that Musk demanded a construction speed faster than Oracle's expectation.
Author of this article: Li Dan.
The weather is good today The weather is good today.
The billion-dollar server rental agreement between xAI, an AI startup owned by Musk, and Oracle may fall through.
During the midday trading session on July 9th, local time in the US, several sources who participated in the negotiations stated that xAI and Oracle have terminated negotiations on expanding the scope of their agreement to lease Nvidia AI chips from Oracle. xAI is now looking to buy chips to create its own AI datacenter in Tennessee and Oracle is not involved in this project. xAI has previously rented 160,000 Nvidia chips from Oracle, and even if previous negotiations have failed, the existing lease agreement will continue to be executed.
Some netizens commented that Oracle missed an opportunity.
After the negotiations with xAI broke down, Oracle's stock price fell rapidly during the midday trading on Tuesday, July 9th, dropping to a daily low of $138 and falling more than 4.8% during the day, before narrowing to less than 3% and ultimately ending the seven-day rally, falling from the closing high set on Monday.
This latest news on Tuesday means that Oracle has lost a $10 billion deal from xAI. In May of this year, the Wall Street Journal reported that xAI was close to reaching an agreement to lease Oracle AI servers for multiple years at a cost of $10 billion, which would make xAI one of Oracle's biggest customers. The size of this transaction, at $10 billion, is similar to those reached by Oracle's cloud business competitors such as OpenAI, Anthropic, Microsoft, Amazon, and Google.
At the time, it was reported that xAI was attempting to complete a $6 billion equity financing round to help cover the cost of cloud computing services. Although the deal with Oracle has not yet been finalized, it suggests that xAI will need more funding to support further development of its AI projects.
On Tuesday of this week, insiders revealed that xAI and Oracle were discussing multi-year leasing of Nvidia chips for xAI's supercomputing plan, but the negotiations have stalled due to several issues, including Musk's demand to build the supercomputer faster than Oracle expected and Oracle's concern that xAI's preferred site for the supercomputer cannot receive sufficient electrical utilities.
Oracle and Musk have not yet responded to media inquiries regarding the above reports.
In April, Musk revealed that xAI would need 100,000 Nvidia H100 GPU chips to train the Grok 3.0 AI model, which is currently being trained with approximately 20,000 H100 chips. He hopes to achieve the goal of 100,000 GPUs by the second half of next year, 2025. In May, Musk told investors that xAI plans to connect all of these chips into a large computer, which is what he calls a "supercomputer factory".
Last week, xAI posted on social media, revealing photos of Musk and other xAI employees. The photos showed them in a datacenter that appeared to be filled with servers. The post did not disclose the specific location, but in June, the president of the Greater Memphis Chamber in Tennessee stated that xAI plans to build a supercomputer at a former Elexesi factory in Memphis. The spokesperson for the Greater Memphis Chamber declined to comment on this.
