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

Sure! Here’s a rewritten version of the article:

—

I stumbled across this thing about Huawei’s CloudMatrix 384 AI cluster. It’s starting to get some love from Chinese tech giants, which is kinda wild, right? They’re pretty hyped about how it stacks up against NVIDIA’s stuff.

So, Huawei’s just going for it, expanding its AI game — like, full throttle. They’re swooping in on NVIDIA’s turf in China. Not too long ago, I read they’ve got this CloudMatrix 384 AI cluster, totally built with their own gear, and it’s sorta giving NVIDIA a run for their money over there. It was in the Financial Times or somewhere, saying ten clients jumped on this server bandwagon. That’s huge for Huawei’s dreams in AI, I guess?

Who these tech giants are? No one’s spilling the beans yet. But they’re all big shots in Huawei’s circle. We’ve talked about CloudMatrix 384 before, but just a quick recap — it’s supposedly standing toe-to-toe with NVIDIA’s beast, the GB200 NVL72. It kinda proves China’s doing fine on its own tech-wise.

And those specs, though! CloudMatrix 384, or CM384, packs in 384 Ascend 910C chips. There’s this funky “all-to-all topology” setup or something. Somehow, Huawei crammed in five times more Ascend chips than the GB200. It spits out 300 PetaFLOPS of BF16 computing power. That’s, like, almost twice what the GB200 NVL72 can do. But man, this thing hogs energy, almost four times more than what the GB200 guzzles. Power efficiency? Not its strong suit, let’s be real.

The price tag is no joke either — $8 million for just one CM384 cluster, which is almost triple what NVIDIA asks for a GB200 NVL72. Seems like Huawei’s aiming for a homegrown powerhouse, not a budget-friendly machine. Makes you think, huh?

Anyway, I’ve rambled enough about it. Just something about this story grabbed me — maybe it’s how tech competition is heating up globally or whatever. Who knows.