Alibaba has unveiled the Zhenwu V900, an AI processor designed by its T-Head subsidiary and presented as three times faster than the previous Zhenwu M890. The chip supports training and inference from FP32 down to FP4, includes 216GB of memory and 1,200GB/s of inter-chip bandwidth, and is scheduled for mass production and commercial release in the first quarter of 2027. Alibaba says a cluster could contain up to 500,000 chips.
The announcement is part of a broader computing push. Alibaba is training Qwen 4 and planning Qwen 4.5 and Qwen 5 models with five to ten trillion parameters, while targeting 20GW of global data-center capacity by 2032. The company has committed more than $53 billion to AI expansion over three years, as US export controls restrict access to leading foreign accelerators. The strategy matters because Alibaba is trying to build domestic infrastructure, models and chips at scale while China faces manufacturing and supply constraints.
https://thenextweb.com/news/alibaba-zhenwu-v900-ai-chip-20gw-data-centres-qwen-5
oxhak / Ox HaK
npub1sx…dz7z7
2026-09-23 00:06:10 UTC
Author Public Key
npub1sxexewvzysc3affq4yzzh7w8e3udyujap2vlj7t6lkdg5dvhp24q4dz7z7Published at
2026-09-23 00:06:10 UTCEvent JSON
{
"id": "dd2093b28e160dc8e2be424ae9e6c923ae8f740977433b6171a7929ce7c07727",
"pubkey": "81b26cb98224311ea520a9042bf9c7cc78d2725d0a99f9797afd9a8a35970aaa",
"created_at": 1790121970,
"kind": 1,
"tags": [],
"content": "Alibaba has unveiled the Zhenwu V900, an AI processor designed by its T-Head subsidiary and presented as three times faster than the previous Zhenwu M890. The chip supports training and inference from FP32 down to FP4, includes 216GB of memory and 1,200GB/s of inter-chip bandwidth, and is scheduled for mass production and commercial release in the first quarter of 2027. Alibaba says a cluster could contain up to 500,000 chips.\n\nThe announcement is part of a broader computing push. Alibaba is training Qwen 4 and planning Qwen 4.5 and Qwen 5 models with five to ten trillion parameters, while targeting 20GW of global data-center capacity by 2032. The company has committed more than $53 billion to AI expansion over three years, as US export controls restrict access to leading foreign accelerators. The strategy matters because Alibaba is trying to build domestic infrastructure, models and chips at scale while China faces manufacturing and supply constraints.\n\nhttps://thenextweb.com/news/alibaba-zhenwu-v900-ai-chip-20gw-data-centres-qwen-5",
"sig": "ff808a6907c4b5ac770de9237712c5915b3387e1ca729abe293e217788ac64f58ecb80a7a88416faa236d0e32b26c9328ce4293f4f94508f7a4e001bcf607956"
}