Oracle is in talks with Meta on a cloud computing deal worth about $20 billion. The multiyear agreement would supply computing power for training and deploying artificial intelligence models. People familiar with the talks said the total may rise, and terms could change before a final agreement.
Talks center on AI training and deployment
The potential deal would make Oracle a provider for Meta’s AI work. It would cover the compute needed to train models and then run them in production. The discussions are private, and the people who described them were not named.
The amount is about $20 billion. The commitment could increase if the companies expand the scope. No final terms are set.
Oracle’s role as an infrastructure provider
The talks add evidence that Oracle has become a significant infrastructure provider. The company has pursued large cloud customers for AI workloads. These include training runs that need vast compute, and deployments that need reliable performance.
According to Bloomberg, the discussions focus on a multiyear supply of compute. The report said the details may change before a final deal.
Details remain private and may change
People familiar with the talks said the discussions are ongoing. They asked not to be named because the talks are private. They also said the total commitment may increase.
The companies have not announced a final agreement. The deal would give Meta more compute for AI efforts. It would also strengthen Oracle’s position in cloud infrastructure.
The report did not list exact timelines. It did not describe specific data centers, chip types, or regions. It only said the deal is multiyear and focuses on AI training and deployment.
Bloomberg reported the talks on September 19, 2025. The story noted that other terms could still change before any signing. The amount, about $20 billion, is a current figure from the talks.
If agreed, the deal would link Meta’s AI needs to Oracle’s cloud. It would support model training and live services. The size of the commitment suggests large and sustained compute demand.