Huawei Delivers CloudMatrix 384 AI Clusters to Chinese Customers at Triple the Cost of NVIDIA’s GB200 NVL72 System

Sure thing, here goes:

—

So, Huawei’s got this wild new AI thing going on, the CloudMatrix 384. It’s like they’re playing this sneaky game against NVIDIA, especially in China. Like, they’re not kidding around. They’ve been shaking things up lately, and now it’s catching some big eyes. Ten bigwigs out there have already jumped on the bandwagon, or so they say. Makes you wonder who’s in that mysterious group, right?

You ever just feel like Huawei’s pushing all the right buttons in the AI world? Yep, they’re really doing their own thing now. This CloudMatrix 384, it’s their baby, totally built in-house. Kind of like their way of waving a flag saying, “Hey, NVIDIA, watch your back.” Who’s depending on who now? Apparently not China, if you get what I mean.

But wait, here’s a fun fact—these things are packed with 384 Ascend 910C chips. They wired them up in some complex, crisscross manner. I don’t know, something about “all-to-all topology.” Fancy, right? Anyway, they’re saying it can crank out 300 PetaFLOPS of BF16 computing. That’s a lot! Double what NVIDIA’s doing, supposedly. Though, rumors say it’s a real juice guzzler, like more than three times the power you’d need for NVIDIA’s gadgets. Ouch.

Here’s the kicker, though: if you’ve got a spare $8 million lying around, you can snag one of these clusters. That’s like three times the cost of the NVIDIA setup. Not exactly chump change. But I guess that’s not their game—they’re not looking to be cheap. They’re making waves their own way, showing off what they can whip up right from their own backyard, sticking it to the Western big dogs.

Crazy times we’re living in, huh? Huawei’s making moves, and everyone’s just waiting to see where the chips fall next.

—