Huawei’s CloudMatrix 384 AI cluster is catching the eyes of Chinese tech heavyweights, sticking it to NVIDIA in wild ways. I mean, why not? It’s like they’re on a mission, refusing to sit back and let NVIDIA hog the AI spotlight. So, they whipped up this newfangled CloudMatrix 384 with all their homegrown tech bits and bobs. It’s doing the rounds, making folks think twice about shelling out for NVIDIA. Financial Times chimed in saying, no lie, ten big players are already onboard with Huawei’s fancy server. Quite the plot twist in the AI playground, right?
So, rumor has it these top-tier Chinese companies have cozied up to the CloudMatrix 384 already. Names are still hush-hush, but these are apparently folks who are all-in on Huawei’s stuff. Now, if you need a cheat sheet on this, think of the CloudMatrix 384 like a gutsy champ going toe-to-toe with NVIDIA’s heavyweight, the GB200 NVL72. It’s like China calling dibs on doing its own thing tech-wise.
Oh, and there’s a picture or something, but honestly, who needs it? Anyway, let’s look at specs. The CloudMatrix has this wild setup — 384 Ascend 910C chips all playing nicely in a web-like configuration. Like, Huawei just threw down more of these Ascend chips than you can shake a stick at, totally outnumbering NVIDIA’s GB200. In terms of raw horsepower, it belts out 300 PetaFLOPS of BF16, almost giving NVIDIA’s best a run for its money! But yeah, here’s the catch—it guzzles power like there’s no tomorrow, 3.9 times more than the GB200. Yikes on the efficiency front, huh?
And dig this—$8 million per CloudMatrix 384 cluster! Almost triple NVIDIA’s GB200 NVL72. It’s not about cheap thrills, apparently. Huawei’s playing the long game with this homespun contender against Western tech titans. Whether it’s a hit or just a buzz, who knows? Might as well grab popcorn and watch the sparks fly.