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

Uh, okay, so Huawei’s got this new AI thing—CloudMatrix 384—making waves with Chinese tech heavyweights. Apparently, it’s got NVIDIA in a bit of a conundrum. You know, like when your favorite pair of socks finally gets that hole and now you’ve got to consider the new ones you didn’t want to buy. Anyway, this AI cluster is all homegrown — Huawei’s brainchild, and it’s causing some buzz, sounding like a bug zapper with the precision of a Swiss watch.

So, right, let’s rewind a bit. Financial Times, or whoever, claims ten major folks have grabbed onto this AI cluster like it’s the last piece of chocolate cake at a party. No names, though. Probably some secret society stuff. But if you put on your detective hat, you can guess they’re the big kids at the tech playground in China. This CloudMatrix thing, it’s like, not just any AI server—it’s like the ninja version of NVIDIA’s GB200 NVL72, less the stealth because apparently, it eats power for breakfast. Like, a lot.

Picture this: 384 Ascend 910C chips all pals, hand-in-hand in some weird tech friendship circle. That’s how they built it, with this “all-to-all topology.” Sounds fancy, right? It’s got this beefy 300 PetaFLOPS of computing power. But, it’ll guzzle electricity like it’s happy hour—or that’s what it seems. Three-point-nine times more energy than NVIDIA’s deal. Ouch. The power meter’s screaming.

Oh! And get this, a single one of these babies costs $8 mil. Yep, triple NVIDIA’s GB200. So, clearly, it’s not about being the guy with cheap plastic trophies but rather making something solidly in-house that says, “Look Ma, no Western hands!” Yeah, a bit like the old days, when you learned to ride a bike without the training wheels. Exciting, huh?