- AMD targets Nvidia’s Blackwell with upcoming Instinct MI355X accelerator
- Oracle plans massive 30,000-unit MI355X cluster for high-performance AI workloads
- That’s in addition to Stargate, Oracle’s 64,000-GPU Nvidia GB200 cluster
While AI darling Nvidia continues to dominate the AI accelerator market, with a share of over 90%, its closest rival, AMD, is hoping to challenge the Blackwell lineup with its new Instinct MI355X series of GPUs.
The MI355X, now expected to arrive by mid-2025, is manufactured on TSMC’s 3nm node and built on AMD’s new CDNA 4 architecture. It will feature 288GB of HBM3E memory, bandwidth of up to 8TB/sec, and support for FP6 and FP4 low-precision computing, positioning it as a strong rival to Nvidia’s Blackwell B100 and B200.
In 2024, we reported on a number of big wins for AMD, which included shipping thousands of its MI300X AI accelerators to Vultr, a leading privately-held cloud computing platform, and to Oracle. Now, the latter has announced plans to build a cluster of 30,000 MI355X AI accelerators.
Stargate
This latest news was revealed during Oracle’s recent Q2 2025 earnings call, where Larry Ellison, Chairman and Chief Technology Officer, told investors, “In Q3, we signed a multi-billion dollar contract with AMD to build a cluster of 30,000 of their latest MI355X GPUs.”
Although he didn’t go into further detail beyond that, Ellison did talk about Project Stargate, saying, “We are in the process of building a gigantic 64,000 GPU liquid-cooled Nvidia GB200 cluster for AI training.”
He later added, “Stargate looks to be the biggest AI training project out there, and we expect that will allow us to grow our RPO even higher in the coming quarters. And we do expect our first large Stargate contract fairly soon.”
When questioned further about Stargate by a Deutsche Bank analyst, Ellison gave a reply that could just as easily apply to the cluster of MI355X AI accelerators Oracle is planning to build.
“The capability we have is to build these huge AI clusters with technology that actually runs faster and more economically than our competitors. So it really is a technology advantage we have over them. If you run faster and you pay by the hour, you cost less. So that technology advantage translates to an economic advantage which allows us to win a lot of these huge deals,” he said.
Ellison also touched on Oracle’s data center strategy, saying, “So, we can start our data centers smaller than our competitors and then we grow based on demand. Building these data centers is expensive, and they’re really expensive if they’re not full or at least half full. So we tend to start small and then add capacity as demand arises.”
Leave a comment