Baidu’s CEO Robin Li revealed that the company has successfully deployed over 30,000 of its third-generation Kunlun P800 AI chips. These chips are capable of training complex AI models similar to DeepSeek. According to Li, this massive deployment enables Baidu to train models with hundreds of billions of parameters, or hundreds of fine-tuned models with billions of parameters each, simultaneously.
The Kunlun P800 chips aren’t just limited to Baidu’s internal use; several Chinese banks and internet companies are already leveraging them for their AI needs.
In a related announcement, Li unveiled Ernie 4.5 Turbo, Baidu’s latest AI model, which claims to match the capabilities of the most advanced AI models available. Ernie 4.5 Turbo excels in tasks such as coding and language comprehension. Additionally, Baidu introduced Ernie X1 Turbo, a model designed for reasoning tasks.
Baidu plans to integrate its AI capabilities across its range of apps, from cloud storage to its content platform, Baidu Wenku. Li emphasized that while AI models are crucial, it’s the applications that ultimately drive real-world impact. “Apps rule the world,” Li stated. “Without apps, models and chips are meaningless.”