Oracle Rumored to Purchase and Deploy Tens of Thousands of AMD MI355X AI Accelerator Chips

AMD’s Instinct MI355X accelerator is the company’s next-generation flagship product in the AI domain, and tech giant Oracle recently announced itself as one of its key customers. According to Oracle’s latest earnings report, the company has procured 30,000 MI355X accelerators to build a massive AI computing cluster. Oracle co-founder Larry Ellison revealed: “In the third quarter, we reached a multi-billion-dollar deal with AMD to deploy 30,000 of the latest MI355X GPUs, creating a powerful AI computing capability.”

The MI355X is a data center GPU built on AMD’s new CDNA4 architecture and manufactured using TSMC’s advanced 3-nanometer process. Optimized specifically for AI workloads, its performance is impressive. It delivers 2.3 petaflops of FP16 computing power and boosts FP8 performance to 4.6 petaflops—a roughly 77% improvement over the previous MI300X series. Even more striking is the MI355X’s introduction of support for FP4 and FP6 low-precision numerical formats, pushing its FP4 computing power to a staggering 9.2 petaflops. This low-precision computing advantage significantly enhances operational efficiency while reducing memory usage, making it ideal for the training and inference needs of modern large-scale AI models.

In terms of memory, the MI355X is equipped with 288 GB of HBM3E memory spread across eight stacks, delivering a total bandwidth of 8 TB/s. This is a clear upgrade over the MI325X’s 256 GB and 6 TB/s, positioning it as a direct competitor to Nvidia’s Blackwell series in both memory capacity and bandwidth. The adoption of HBM3E technology ensures efficient execution of data-intensive AI tasks, providing robust support for hyperscale computing.

Oracle plans to leverage this cluster to advance its cloud services and AI product offerings. Ellison noted that four major cybersecurity firms—CrowdStrike, Cyber Reason, Newfold Digital, and Palo Alto Networks—have chosen to migrate to Oracle Cloud. Additionally, Oracle has launched a new product called the AI Data Platform, designed to help its vast database customer base analyze data using the latest AI models. This platform integrates cutting-edge AI technologies from OpenAI, xAI, and Meta, and utilizes the vector capabilities of Oracle Database 23 to automatically convert customers’ existing data into formats AI can understand. This functionality enables businesses to quickly extract value from their data while ensuring privacy and security.

The production and market deployment timeline for the MI355X is set, with AMD slated to begin shipping in the second half of 2025. This schedule aligns with AMD’s annual data center GPU update cycle and provides Oracle ample time to prepare for its data center expansion. Reports indicate that Oracle is doubling down on its cloud infrastructure investments, aiming to double its data center capacity by the end of 2025 to meet the growing demand for AI and cloud computing. By then, the cluster of 30,000 MI355X GPUs will serve as the core powerhouse of Oracle Cloud, enhancing its competitiveness in AI workloads.

AMD is accelerating its efforts in the AI accelerator space. Beyond the MI355X, the company plans to launch the MI400 series, based on the CDNA “Next” architecture, in 2026. The MI355X’s chipset design continues AMD’s signature multi-chiplet packaging strategy, achieving high performance through the collaboration of multiple compute dies. This approach not only improves manufacturing efficiency but also reduces costs, giving AMD greater flexibility in its competition with Nvidia.

Notably, the MI355X’s performance metrics closely rival Nvidia’s Blackwell B200, which also offers 9 petaflops of FP4 computing power. However, the MI355X’s 288 GB of HBM3E memory surpasses the B200’s 192 GB by 50%, providing more headroom for memory-intensive applications. Furthermore, its 8 TB/s bandwidth matches the B200, ensuring efficient data transfer. This hardware comparison highlights AMD’s strategy of countering Nvidia with larger memory capacity and equivalent bandwidth, aiming to capture a bigger share of the AI accelerator market.

Oracle’s procurement decision reflects broader industry trends: surging demand for AI computing power and cloud service providers racing to gain a competitive edge through hardware upgrades. The deployment of the MI355X will enable Oracle to support larger-scale AI models, such as language models and multimodal systems with hundreds of billions or even trillions of parameters. This capability is particularly critical for customers in finance, healthcare, and security, who need to process vast datasets and generate real-time insights.

With the MI355X set to hit the market in full force in the second half of 2025, the collaboration between AMD and Oracle will undoubtedly inject new energy into the development of AI technology. This accelerator represents not only a leap in hardware performance but also a glimpse into the future trend of deep integration between cloud computing and AI. For tech enthusiasts, this hardware-driven AI revolution is one to watch closely.