Okay, let’s dive into this — another day, another tech giant making waves. So, Huawei’s got this thing, the CloudMatrix 384 AI cluster, and everyone’s talking about it. Chinese tech bigwigs are supposedly all over it, especially since it’s giving NVIDIA a run for its money — at least over in China anyway.
Huawei’s Shiny New Toy: Threatening NVIDIA’s Crown? Especially In Its Own Backyard…
Huawei’s not slowing down, not for a second. They’re expanding their AI lineup like there’s no tomorrow. And guess what, they’re taking advantage of NVIDIA’s shaky spot over in the Chinese market. We heard about their new CloudMatrix 384 cluster, built with their own stuff, and it’s supposedly good enough to give NVIDIA a headache domestically. Oh, and just a little nugget from the Financial Times — ten clients have picked up the CloudMatrix 384. Huge for Huawei, right?
Now, who are these mysterious adopters? No one’s telling. But they’re core customers of Huawei, so there’s that. We’ve talked about CloudMatrix 384 before, haven’t we? Anyway, just to recap: it’s said to stand toe-to-toe with NVIDIA’s GB200 NVL72, suggesting China’s kind of doing its own thing now, tech-wise.
Let’s talk specs. The CloudMatrix 384 packs in 384 Ascend 910C chips, all cozy in an "all-to-all topology." Basically, Huawei crammed in five times more Ascend chips than NVIDIA did in their GB200. It cranks out 300 PetaFLOPS BF16 computing, which is nearly double what GB200 offers. But here’s the kicker, it guzzles 3.9 times more power. Eco-friendly? Nah — maybe not its strong suit, huh?
Price tag time: $8 million for each CloudMatrix 384 cluster. That’s close to thrice what NVIDIA’s GB200 NVL72 costs. So clearly, Huawei’s not aiming for budget-friendly here, but rather a homegrown powerhouse to match up against the Western big shots.
