Elon Musk’s AI startup xAI and Oracle’s large-scale server rental negotiations have fallen through. Because of this, Oracle will provide 100,000 GPUs to Microsoft (MS), which can likely be used to develop OpenAI’s models.
The Information reported on the ninth (local time) that xAI had been in talks to lease Oracle cloud servers for several years for a fee of $10 billion (roughly 14 trillion won), however the deal fell through with none results.
It’s reported that the negotiations were halted resulting from unreasonable demands from xAI. CEO Musk requested that the information center be built faster than the schedule presented by Oracle.
Oracle also reportedly expressed concerns about power supply issues in Memphis, Tennessee, where xAI plans to construct an information center.
Ultimately, CEO Musk revealed via X (Twitter) that “xAI has been training its chatbot ‘Grok2’ using 24,000 of Oracle’s NVIDIA ‘H100’ GPUs and is ready to launch next month, but is now constructing its own supercomputer utilizing 100,000 H100 GPUs.”
He continued, “Speed ​​is vital to us,” and “When speed determines your destiny, it’s best to not sit within the back seat but take the wheel yourself.”
So xAI will construct its own AI data center in Memphis, Tennessee, using Nvidia chips supplied by Dell and Supermicro.
Meanwhile, as negotiations with xAI fell through, Oracle is alleged to have agreed to provide 100,000 H100 GPUs to MS that it had originally intended to offer to xAI.
Last month, Oracle announced that it had signed a lease agreement with Microsoft to offer Nvidia GPU-based servers to OpenAI. For this contract, it converted a supercomputer that had been prepared for xAI.
In the long run, CEO Musk seemed to be helping his arch-rival by providing OpenAI with the GPU capability he had secured for xAI.
Reporter Park Chan cpark@aitimes.com
