Overview

• Tencent confirmed it has a significant stockpile of high-end GPUs, sufficient to train AI models for several more generations.
• The company has developed more efficient AI training techniques that reduce reliance on large GPU clusters while maintaining strong performance.
• Tencent is prioritizing GPU allocation for high-return applications, such as advertising and content recommendation systems.
• Exploration of alternative hardware, including domestic chips, ASICs, and smaller GPUs, is underway to diversify and mitigate reliance on US exports.
• CEO Pony Ma announced that GPU rentals for Tencent’s cloud business are being deprioritized due to supply constraints.