Huawei Delivers CloudMatrix 384 AI Clusters to Chinese Clients; Costs Triple NVIDIA’s GB200 NVL72 System

Sure, let’s dive into this Huawei vs. NVIDIA thing with a touch of chaotic human rambling.

So, get this — Huawei has this new toy, the CloudMatrix 384 AI cluster, and apparently, the big tech players in China are pretty stoked about it. Like, they’ve been chatting about it and saying it’s giving NVIDIA a run for their money. And honestly, who would’ve thought? I mean, I can barely get through breakfast without spilling coffee, but these folks are out there juggling server clusters like it’s nothing.

The Deal with Huawei’s CloudMatrix

Huawei’s been on a roll lately, especially in the AI world — and they’re not slowing down anytime soon. They’ve been poking at NVIDIA’s soft spots in China. It’s almost like they’re playing chess, but with servers. Recently, the CloudMatrix 384 AI behemoth got unveiled, which they cooked up entirely in-house. I guess that means they’ve got nerd-level bragging rights because apparently, it’s giving NVIDIA’s top dog servers a headache. The Financial Times dropped a scoop saying a bunch of big names have already started snatching them up, though nobody’s spilling exactly who. It’s like a secret AI club or something. This is huge for Huawei’s goal of being the Big Cheese in AI!

Over here, people say this cluster has squeezed into some data center portfolios. Kinda like how I try to squeeze my sneakers in a packed closet. Anyway, these unnamed tech titans are long-time Huawei fans. We’ve jabbered about the CloudMatrix 384 before, but here’s a crash course: Huawei has built this little marvel to stand toe-to-toe with NVIDIA’s mega server, the GB200 NVL72. Apparently, China’s gearing up to handle its computational biz in-house now — that’s what some folks say, anyway.

There’s a chart or something that’s supposed to explain it all — but let’s be real, charts never make sense until you’ve had two cups of coffee.

The nitty-gritty? Uh, this thing’s got 384 of Huawei’s Ascend chips, all meshed together like an intricate spider web, or a plate of tangled spaghetti. They’ve managed to cram in five times the Ascend chips as NVIDIA’s model, which sounds wild, right? Supposedly, the CloudMatrix churns out a mind-boggling 300 PetaFLOPS of BF16 juice. I don’t even know what that really means, but it sounds cool. On the down side, it sucks power like nobody’s business — apparently, it’s a real electricity guzzler compared to NVIDIA. Makes me wonder if we should start buying stock in power companies.

Now, here’s something that’ll make your bank account sweat: One of these clusters costs about $8 million. That’s, like, triple NVIDIA’s option! Clearly, Huawei isn’t trying to be the budget-friendly choice. Maybe the idea is to flex those in-house muscles and show they can hang with the Western giants. Or, maybe it’s just like a really expensive collector’s item for techies.

Anyway, it seems like Huawei’s making big plays, and who knows where this tech tangle will lead next? For now, let’s just sit back and see how this high-stakes chess match unfolds.