Last mentioned: Mar 21, 2026
Huawei Technologies has launched the Atlas 350 accelerator card, powered by the Ascend 950PR chip, claiming a 2.8x performance lead over Nvidia’s H20 in AI inference tasks. The launch signals a major step in China's semiconductor self-sufficiency as the industry transitions toward agentic AI workloads.
Huawei has unveiled the Atlas 350 AI accelerator card, powered by the Ascend 950PR chip, claiming performance nearly triple that of Nvidia’s H20 in inference tasks. The launch signals Huawei's aggressive push into agentic AI infrastructure as it seeks to bypass US trade restrictions and dominate the domestic AI market.
Nvidia has unveiled the Groq 3 Language Processing Unit (LPU) and the Vera Rubin platform, signaling a strategic shift toward system-level dominance in agentic AI inference. This development widens the hardware gap with Chinese rivals while simultaneously forcing a strategic pivot toward vertical, smaller-scale AI models in the mainland market.
Nvidia has unveiled the Groq 3 LPU and Vera Rubin platform, shifting its focus toward 'agentic AI' and integrated 'AI factories.' This move widens the competitive gap with Chinese chipmakers, forcing a strategic pivot toward vertical-specific inference and smaller-scale model deployments.