While Nvidia still reigns over the AI accelerator world—clutching a near-mythical 90% market share—its long-time rival AMD is sharpening its blades. And with the new Instinct MI355X series, it isn’t just aiming to compete—it’s gunning for a shake-up.
Expected to land by mid-2025, the MI355X is anything but ordinary. Built on TSMC’s cutting-edge 3nm node and AMD’s shiny new CDNA 4 architecture, this GPU is sculpted for power. We’re talking 288GB of HBM3E memory, a mind-bending 8TB/sec of bandwidth, and support for FP6 and FP4 low-precision formats—ideal for the data-hungry, cost-sensitive world of AI training. Sounds like it could rattle Nvidia’s Blackwell lineup, doesn’t it?
But AMD’s not just flexing specs—it’s making moves. Last year, it scored wins with Vultr and Oracle, shipping thousands of MI300X chips. Now, Oracle is doubling down. In its recent Q2 2025 earnings call, CTO Larry Ellison casually dropped a bombshell: a 30,000-GPU cluster of MI355Xs is in the works. That’s not a typo—thirty thousand.
Take a moment and just let that number sink in.
Ellison described the deal as a multi-billion dollar partnership, signaling AMD’s rising influence in hyperscale AI. And while he didn’t unpack the full vision, what he did reveal added a layer of intrigue. He teased Project Stargate—a colossal 64,000-GPU Nvidia GB200 cluster being built for AI training. His tone? Confident. Ambitious. Maybe even a little smug.
And here’s the kicker—Ellison didn’t draw hard lines between Stargate and AMD’s own cluster. In fact, when asked about Oracle’s AI infrastructure strategy, he hinted at something bigger: “We can build massive AI clusters that run faster and cheaper than our competitors. If it’s faster, it costs less. That’s our edge.” It’s not just about hardware. It’s about efficiency. Economics. Momentum.
Of course, building out this kind of scale doesn’t come cheap. Or easy. Ellison spoke candidly about Oracle’s data center approach—start small, scale up with demand. It’s pragmatic, even cautious. “Empty data centers are expensive,” he said. But once they fill up? That’s where the magic happens.
This bold move clearly shows that AMD is no longer just nibbling at Nvidia’s heels. It’s prepared for battle and this time, a long drawn one, until of course, only one remains standing tall over the other. With specs that dazzle, partners with deep pockets, and an ecosystem hungry for alternatives, the MI355X might just be the dark horse of the AI race.
And who knows? Maybe we need to call an ambulance, but for Nvidia this time around.
