Big Red gets cozy with Google to offer TPUs to AI customers

OCI customers are poised to gain a significant array of new choices for their infrastructure. Big Red is hosting an event in Las Vegas this week to detail its latest AI investments, placing a major emphasis on an expanded alliance with OpenAI and plans to broaden its chip selection beyond Nvidia by incorporating Google's Tensor Processing Units (TPUs) into OCI.

Oracle states that its agreement to integrate Google TPUs into OCI will establish it as the first hyperscale cloud platform outside of Google's own services to offer these specific chips. These TPUs are scheduled to become selectable within OCI starting in 2027, with capacity slated to increase across several years.

While the precise volume of Google TPUs for the coming year and beyond remains unstated, Oracle appears keen on lessening its deep reliance on Nvidia, with whom it has secured contracts worth tens of billions of dollars for equipping its data centers, including the reportedly challenging Texas project alongside the GPU maker and OpenAI.

"Different AI workloads have different compute needs," stated OCI president Mahesh Thiagarajan. "Bringing Google TPUs to OCI gives customers … the flexibility to choose the right infrastructure for the performance, throughput, and economics of each AI workload."

The company is not abandoning its OpenAI cooperation, having announced new ChatGPT integrations on Tuesday alongside two new offerings that utilize models from the House of Altman. One, Oracle AI Workspace, is described as a "single user experience that brings OpenAI frontier models together with Oracle applications, enterprise data, and cloud services" via ChatGPT plugins that effectively make Oracle’s current business platforms accessible through ChatGPT.

"The Oracle AI Workspace will make the breadth of Oracle’s capabilities agent-ready and accessible to ChatGPT users," the firm clarified. "People can describe business goals and these plugins in ChatGPT will bring together the right Oracle tools and information to help them accomplish it."

In addition to granting OpenAI-powered tools access to select corporate data, Oracle anticipates customers will trust the models to aid in cybersecurity with its second new product, Oracle AI Security Mesh, which is detailed as "a managed platform that combines frontier AI models for cybersecurity … with Oracle's proven security expertise." This mesh will employ OpenAI cybersecurity models to identify weaknesses, test exploitability, and suggest fixes within a company's authorized tech stack.

Furthermore, Oracle confirmed a deal with Snowflake to introduce its AI data cloud as a native service within OCI. This new collaboration enables Oracle customers to procure and utilize Snowflake directly from Oracle, with Oracle managing all associated billing, discounts, and pricing. Oracle Universal Credits will also be usable on Snowflake within OCI.

Even as Oracle diversifies its OCI chip portfolio, it also unveiled new Nvidia hardware in Las Vegas. Big Red announced the general availability of Nvidia Vera Rubin NVL72 systems and Nvidia Vera CPUs on Tuesday, alongside plans to introduce HGX Rubin NVL8 systems, though a timeline for their arrival was not provided. Thiagarajan commented, "By combining NVIDIA Vera Rubin GPUs and Vera CPUs with Oracle Acceleron networking and AI-optimized data centers, we’re helping customers keep GPUs productive, secure, and scale efficiently as their needs grow."