Tapping public markets will help ease some of the funding pressure for Enflame, founded by Zhao Lidong and Zhang Yalin, two former AMD employees. The company's net loss widened year-on-year by more than a third to 444 million yuan ($66 million) in the three months to March. But its more crucial support probably comes from having Tencent, a mobile messaging-to-gaming behemoth, as a top customer and its 20% owner.
The eight-year-old upstart has a steep hill to climb: unlike Nvidia and domestic peers that focus on general-purpose GPUs — currently the industry's preferred AI chip — Enflame is betting on specialised and more power-efficient processors. These types of customised silicon are starting to take off, particularly for inference, where trained models process prompts and generate a response. Still, Enflame's strategy shows how the little dragons are tackling Nvidia's twin advantages.
In hardware, Chinese challengers are lasering in on the fast-growing market for inference chips, which does not require as much computing power as those used to train models. And crucially, in software, Enflame is developing its own programming platform to rival Nvidia's dominant CUDA ecosystem, which hosts many apps, frameworks and AI software that have become industry standard. Nvidia has relied on CUDA to lock in customers. Similarly, the $36 billion Moore Threads, founded by former Nvidia executive Zhang Jianzhong, boasts a custom system that helps developers migrate code written using CUDA to its own language.
That Nvidia's market share in China has fallen from a near monopoly a few years ago to an estimated 55% today underscores China's success in cultivating domestic alternatives in areas outside of training and how the little dragons will keep taking bites out of their U.S. rival.
NVDA•
