OpenAI plans Google AI chip purchases, media report says
The ChatGPT maker previously began renting Google's chips. Financial terms of the planned purchases were not disclosed.
Oracle broadens AI chip lineup as it expands Nvidia agreement
Google TPUs will join Oracle’s cloud offerings for the first time, while chips from SambaNova Systems and Cerebras Systems are also planned.
Oracle announced on October 27 that it would increase its use of Nvidia AI servers and widen the selection of rival processors available through its cloud. Oracle and Nvidia said the expansion would bring deployments of Nvidia’s newest Vera Rubin systems to a scale of multiple gigawatts.
Financial terms were not disclosed. Nvidia CEO Jensen Huang has previously put the potential revenue from Vera Rubin systems at up to $40 billion per gigawatt for Nvidia.
The broader lineup will include tensor processing unit chips from Alphabet’s Google, which Oracle has not offered before. Oracle also intends to add processors from Intel-backed SambaNova Systems and Cerebras Systems. Both companies produce chips for inference, the process of responding to users’ requests.
Oracle said its own Acceleron networking chips would take the place of some Nvidia networking components in its Nvidia-based systems. Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, said AI workloads have varying computing requirements and that offering alternatives gives customers flexibility.
Oracle gave no details on financing the purchases. In an earnings call the month before the announcement, it said customers were increasingly covering chip costs. It left its capital spending target unchanged at about $90 billion to $95 billion for fiscal 2027.