[News] China’s Moore Threads Unveils Huashan AI Chip, Reportedly Takes Aim at NVIDIA’s Hopper

Hot on the heels of the U.S. greenlighting H200 exports to China, the country’s so-called “little NVIDIA,” Moore Threads, has unveiled a new product lineup, headlined by the AI training and inference chip Huashan, aiming to challenge Team Green’s Hopper series, according to the South China Morning Post.

The report notes that the Huashan chip, along with Lushan, a new GPU targeting high-performance graphics, is slated for mass production in 2026.

Citing founder and chairman James Zhang Jianzhong at a developer conference last Saturday, the South China Morning Post reports that Huashan—named after an iconic Chinese mountain—could even rival NVIDIA’s latest Blackwell lineup. Zhang said Huashan outperforms Hopper models, including the widely used H100 and H200, in compute power, memory bandwidth, and capacity, though full specifications remain under wraps.

According to Wccftech, the Huashan GPU seems to use a dual-chiplet design and comes with eight HBM sites, allowing for massive memory bandwidth and access. The report also suggests its floating-point performance nears the Blackwell B200, while memory capacity could even surpass it, highlighting Moore Threads’ ambition in the high-end AI GPU space.

Additionally, the new GPUs are expected to offer up to four times the memory capacity, Wccftech reports, noting that with the current MTT S80/S90 chips packing 16 GB of GDDR6 memory, upcoming models could reach as much as 64 GB.

Other Chips Make Their Debut

Meanwhile, Moore Threads also unveiled Lushan, which reportedly delivers up to 15 times the performance for triple-A titles—a term for games with massive budgets and extended development cycles, as per the South China Morning Post.

Both new chips are based on Huagang, Moore Threads’ next-generation GPU architecture, which Zhang said boosts performance by 10 percent over the previous generation, the report adds.

In addition, Moore Threads has unveiled its first intelligent SoC chip—Yangtze, as noted by IT Home.

The Yangtze chip, named after China’s longest river, supports 32GB or 64GB of LPDDR5X memory with bandwidth exceeding 100 GB/s, and delivers up to 50 TOPS of heterogeneous AI compute, IT Home reports, adding it has already been officially announced in the MTT AIBOOK laptops and MTT AI Cube mini PCs.

Read more

• [News] China’s GPU IPO Wave Intensifies: MetaX Surpasses Moore Threads in Retail Investor Interest
• [News] China’s GPU Trio Rise as NVIDIA Retreats — Decoding Moore Threads, MetaX, and Cambricon
• [News] Baidu Reportedly Explores Kunlunxin Spinoff and Listing Amid a Surge in China Chip IPOs

(Photo credit: Moore Thread’s Weixin Official Account)