Nvidia’s AI crown is looking shakier by the minute. AMD just scored a monster deal with Oracle to build a 30,000-strong cluster of its next-gen MI355X accelerators—a direct shot across Blackwell’s bow. Forget baby steps; this is a full-tilt sprint into the AI arms race.
Let’s talk specs, because these numbers are ludicrous. Built on TSMC’s cutting-edge 3nm process and AMD’s CDNA 4 architecture, the MI355X packs 288GB of HBM3E memory, 8TB/sec bandwidth, and support for FP6/FP4 low-precision computing. Translation? It’s not just competing with Nvidia’s B100 and B200—it’s gunning for their lunch money.
But wait, there’s more. Oracle’s Larry Ellison casually dropped this bombshell during the Q2 earnings call, sandwiched between chatter about Project Stargate, their 64,000-GPU Nvidia GB200 behemoth. The audacity! Running dual-track with AMD and Nvidia clusters? That’s not hedging bets—that’s playing 4D chess.
Here’s the kicker: Ellison’s boast about Oracle’s “faster, more economical” tech isn’t just corporate fluff. With AMD’s hardware, Oracle can scale data centers leaner and meaner than rivals. “If you run faster and pay by the hour, you cost less,” he smirked. Mic drop.
What does this mean for the AI war? Nvidia still rules the roost, but AMD’s MI300X wins with Vultr and Oracle prove it’s no longer a footnote. With the MI355X landing mid-2025, the battlefield just got a hell of a lot more interesting.
