Samedia.ai—China is making significant strides in artificial intelligence, with many companies emerging as leaders. One standout is Deepseek AI, a startup gaining attention for its innovative AI research and development. Founded by Liang Wenfeng, a post-80s entrepreneur, Deepseek AI is pushing the boundaries of AI technology.
Deepseek AI’s approach differs from many Chinese companies that focus on commercialization. Instead, it emphasizes driving innovation and exploring AI’s possibilities. This strategy has led to breakthroughs, including the development of a novel MLA (multi-head latent attention) architecture. This new architecture reduces memory usage to 5-13% of the commonly used MHA architecture, making AI models more efficient and effective.
In May, Deepseek AI released an open-source model, DeepSeek V2, with an unprecedented price/performance ratio. This move triggered a price war in the large-model market, with major tech giants like ByteDance, Tencent, Baidu, and Alibaba cutting prices. The competition spurred innovation, forcing companies to improve their models to stay competitive.
CEO Liang Wenfeng’s Insights
In an interview with Waves, CEO Liang Wenfeng explained how DeepSeek V2’s release inadvertently started a price war. “We didn’t mean to become a catfish — we just accidentally became a catfish,” Liang said, referring to the market disruptor role. He was surprised by the sensitivity to pricing and stated that their principle is not to subsidize or make exorbitant profits but to maintain a small profit margin above costs.
Liang noted that Zhipu AI followed by reducing the price of an entry-level product, while ByteDance matched Deepseek’s flagship model price, prompting others to follow. “We never expected anyone would do this at a loss,” Liang remarked, “but it turned into the familiar subsidy-burning logic of the internet era.”
Liang emphasized that their goal was not to poach users but to make AI and APIs accessible and affordable. He also highlighted the importance of starting from the model structure rather than copying current generation architectures, aiming for stronger model capability with limited resources.
Discussing the innovation gap, Liang pointed out that China’s capabilities might have a twofold gap in model structure and training dynamics compared to international levels, requiring twice the computing power to achieve the same results. He emphasized the need for China to contribute to global innovation rather than freeriding on the success of others.
Future Prospects
Deepseek AI’s innovative approach signals a new era in AI research. Its focus on exploration and open-source models allows for significant breakthroughs and global collaboration. As the AI industry evolves, Deepseek AI is poised to make substantial contributions, pushing the boundaries of what is possible with AI.
The future of AI research looks bright, with Deepseek AI at the forefront. The company’s approach ensures it remains a key player in AI innovation, driving progress and setting new standards in the field.
