Oracle has revealed some of the AI companies using its cloud and AI infrastructure for training models, including Elon Musk's xAI.

xAI will use Oracle Cloud Infrastructure's (OCI) AI infrastructure for the training of its next-generation Grok models and to run inference workloads.

The models will also be offered via OCI's generative AI service.

“Grok 3 represents a leap forward in AI capabilities, and Oracle’s advanced data platform will accelerate its impact on enterprises,” said Jimmy Ba, co-founder, xAI. “This collaboration between xAI and Oracle is set to redefine enterprise-grade AI.”

Oracle has an extensive fleet of GPUs in operation, including its "OCI Superclusters" made up of Nvidia GPUs. The company recently launched a Supercluster with 131,072 Blackwell GPUs, has another cluster of 65,536 Nvidia H200 GPUs, and an H100 Supercluster that can scale to 16,384 GPUs.

Big Red is also planning a supercluster of AMD GPUs, which will have 131,072 of the vendor's new MI355X chips.

While xAI is using OCI for training, the AI company also operates its own data center with a large number of GPUs.

The company's Memphis facility was launched in July 2024, with an initial capacity of 150MW. However, last October, it was reported that the capacity at the data center was set to double, following the installation of an additional 100,000 Nvidia Hopper GPUs to the cluster. It is located on an industrial park at Paul Lowry Road, in southwest Memphis.

xAI is facing a lawsuit as a result of the increase in power capacity, for which the company allegedly did not obtain the necessary permits to legally operate the 26 natural gas turbines that were installed on the site to power its operations.

The company is planning a second data center in Memphis, Tennessee, following the purchase of a one million sq ft (92,903 sqm) site on Tulane Road.

Oracle's many AI customers

In addition to xAI, Oracle has detailed other AI companies using OCI for their model training and inference workloads.

Among the companies announced are Fireworks AI, Hedra, Numenta, and Soniox.

“Among AI innovators, OCI has rapidly become the destination of choice for training and inferencing needs of all sizes,” said Chris Gandolfo, executive vice president, Oracle Cloud Infrastructure and AI. “OCI AI infrastructure delivers ultra high-speed networking, optimized storage, and cutting-edge GPUs that AI companies rely on to power the next wave of innovation.”

Inference platform Fireworks AI is using OCI Compute bare metal instances with Nvidia Hopper GPUs and OCI Compute with AMD MI300X GPUs to serve more than two trillion inference tokens daily.

Similarly, Soniox, an AI company working on foundational AI models for audio, speech, and language comprehension, is also using OCI Compute bare metal instances with Nvidia Hopper GPUs to train its model to recognize and understand speech across 60 languages with low latency and higher accuracy.

“A high-performance infrastructure that provides improved accuracy, speed, and cost efficiency was top-of-mind when selecting a cloud provider to support our growth,” said Klemen Simonic, founder and CEO, Soniox. “OCI gives us access to the latest AI innovations that allow us to push the boundaries of speech recognition and audio understanding while significantly reducing deployment time and operational costs.”