Chinese PCB giant Alibaba Cloud, Zhiyuan leaders call AI era ‘still early’

This conversation between Wang Jian and Huang Tiejun is a useful window into how senior Chinese technologists are framing the moment. Their core message—that we’re still in a messy, early stage of understanding intelligence, and that foundational research belongs to the whole world—pushes back on both AI doom and AI triumphalism. Wang’s analogy that failed multi‑month training runs are now ‘like blowing up a rocket’ also captures how capital‑intensive top‑end experimentation has become, and why only a handful of actors can afford to push the frontier.

For the AGI race, the interesting signal is how they talk about data and cost. Wang emphasizes that the real revolution will be AI’s ability to reinterpret scientific and code data, not just text, and that token prices must fall until they feel as cheap as paper. That implies a long‑term view where AGI‑like systems are woven into scientific discovery itself, not just consumer apps. It also hints at a future where China is less focused on cloning US models and more on using AI to accelerate its own science and industry. If that vision holds, global competition will hinge less on single ‘AGI breakthroughs’ and more on who can afford to keep re‑training models on ever‑richer scientific and industrial datasets.