Submission + - China Testing Domestic Advanced Chipmaking Tool, Plan for AI Clusters (ft.com)
In a related development, Chinese telecommunications giant Huawei announced Thursday new computing systems for powering artificial intelligence with its in-house Ascend chips despite heavy American sanctions against the company. Huawei announced it would roll out three new versions of its Ascend chips through the end of 2028, with the aim to “double compute” capabilities with each year’s release, and build AI superpods and superclusters from the new chips. According to Huawei, the 950PR, to be released early 2026, will feature 128GB of its in-house HBM delivering up to 1.6 TB/s of bandwidth, while the 950DT increases those figures to 144GB and 4 TB/s, but Huawei hasn’t disclosed how its in-house HBM is manufactured, what packaging is used, or which foundry is producing the chip itself. That is followed by Ascend 960 and 970, to be released in 2027 and 2028 respectively. Its new Atlas 950 supernode would support 8,192 Ascend 950 chips, and that the Atlas 950 SuperCluster would use more than 500,000 chips. A more advanced Atlas 960 version, slated for launch in 2027, would support 15,488 Ascend 960 chips per node. The full supercluster would have more than 1 million Ascend chips, according to Huawei. In a speech Thursday, Eric Xu, rotating chairman of Huawei, claimed that its forthcoming Atlas 950 supernode would deliver 6.7 times more computing power than Nvidia’s NVL144 system, also planned for launch next year, and the Atlas 950 supercluster would have 1.3 times the computing power of Elon Musk’s xAI Colossus supercomputer. China on Monday announced it was extending a probe into Nvidia over alleged monopolistic practices and ordered local tech giants to stop tests and orders of the Nvidia ("castrated by the US government") H20 and RTX Pro 6000D chips.