Huawei's New AI Cluster Takes On Nvidia, But At A Higher Cost Following Sanctions

Zinger Key Points

Huawei delivers its CloudMatrix AI chip, taking advantage of Nvidia’s sanctions, despite higher energy costs and maintenance needs.
CloudMatrix 384 aims to rival Nvidia’s AI chips, but Huawei's setup is pricier and more energy-intensive than Nvidia’s efficient systems.
9 Out of the Last 10 Summers this "Power Pattern" Delivered Winners - Get The Details Now.

Huawei Technologies has started delivering its advanced artificial intelligence chip “cluster” to Chinese clients, who ramped up orders after Washington’s semiconductor sanctions cut them off from Nvidia Corp’s NVDA sophisticated technology.

The erstwhile Chinese smartphone giant has sold over 10 sets of CloudMatrix 384, which fuses a large sum of chips, the Financial Times reported Wednesday, citing unnamed sources familiar with the matter.

Nvidia stock is trading lower on Wednesday.

Also Read: Nvidia Powers Next-Gen Adaptive Digital AI Agents For Cisco, Nasdaq, And Meta

Dylan Patel of SemiAnalysis told the Financial Times that Huawei’s CloudMatrix 384 development signifies that China now has an AI system capable of beating Nvidia.

Nvidia projected a $5.5 billion earnings hit after President Donald Trump mandated a license to sell H20 chips to China.

Huawei claimed its CloudMatrix outperformed Nvidia’s NVL72 AI cluster, which consists of 72 of its GB200 chips, the Financial Times cited from a company presentation.

CloudMatrix 384 uses Huawei’s Ascend 910C chips, which underperform Nvidia’s GB200 processors. However, Huawei has used a larger quantity of chips connected by its “super node.”

However, CloudMatrix 384 has several disadvantages compared to Nvidia, industry experts told Financial Times.

CloudMatrix 384 has much higher energy consumption due to the higher number of chips, and Huawei’s software systems require more maintenance from experienced engineers, entailing higher costs than Nvidia.

CloudMatrix 384 sells for about 60 million Chinese yuan ($8.2 million), a set compared to $3 million for Nvidia’s NVL72.

BofA Securities analyst Vivek Arya had highlighted growing concerns that AI data center buildouts might peak in 2025 and then considerably slow down afterward.

However, he also expects that capex from non-CSPs, including emerging cloud vendors, enterprises, and national AI infrastructure projects, could be an additional source of sustaining AI investments.

Arya’s top AI picks were Nvidia and Broadcom Inc AVGO, respective merchant and custom AI silicon leaders.

Price Action: NVDA stock is down 1.23% at $107.68 at the last check Wednesday.

Read Next: