Big Tech Pivot to Embodied AI
Key Questions
What are Alibaba and Tencent doing with their LLMs in robotics?
Alibaba is embedding Qwen3.7-Max with tool-calling capabilities for physical robot control, while Tencent is integrating OpenClaw into Zeroth's M1 robot. This marks a strategic pivot from chatbot-focused AI to embodied AI systems.
What is Alibaba's Qwen-VLA paper about?
Qwen-VLA introduces a unified approach to vision-language-action modeling that works across tasks, environments, and different robot embodiments. It further signals Alibaba's commitment to advancing embodied AI beyond language models alone.
What is the main challenge in scaling embodied AI according to the highlight?
A key constraint is the data bottleneck, with roughly 500k hours of available training data versus 10B hours used for models like GPT-5. China's ongoing expansion of training facilities reflects efforts to address this limitation.
Alibaba and Tencent are embedding their LLMs directly into robots – Alibaba's Qwen3.7-Max with tool-calling for physical control, Tencent's OpenClaw in Zeroth's M1. This signals a major strategic pivot from chatbots to embodied AI, validating the sector's investment thesis. The data bottleneck (500k hours vs GPT-5's 10B hours) is a key constraint, but China's rapid buildout of training facilities shows intent. Alibaba also released Qwen-VLA paper unifying vision-language-action modeling across tasks and embodiments, further demonstrating commitment.