Baidu spent two decades as China's answer to Google. Search was the business. Everything else was secondary. That era has ended. The company is now positioning itself as a full-stack artificial intelligence player with its own silicon, proprietary models, cloud infrastructure, and robotaxi operations. The change is not incremental. It is structural.
CFO Henry He's recent comments to Bloomberg laid out the new logic. Management is focused on maximising token spend, embedding safety into development, and competing globally in robotaxis. Search is no longer the primary business line. It is a foundation that feeds training data into Ernie, Baidu's large language model series, which runs on proprietary compute that Baidu built. The models power Apollo Go, the robotaxi unit. The robotaxi unit generates operational data that improves the models. The stack feeds itself.
This is the Nvidia playbook applied to China. Nvidia built full-stack dominance by controlling chip architecture, software layers, and developer relationships. Every company that wants to build AI needs Nvidia's chips. Every company building at scale needs to understand Nvidia's software stack. Nvidia extracts value at every layer. Baidu is attempting the same thing domestically.
The rationale is sovereignty. China cannot rely on Nvidia forever. US export controls are tightening. Baidu needs its own chip capability to guarantee supply. It needs proprietary models to differentiate and avoid dependency on OpenAI or Anthropic. It needs proprietary cloud infrastructure to control cost and lock in customers. And it needs a robotaxi unit to prove the entire stack works at scale in the real world.
The search business fits into this as a moat and a training ground. Search generates hundreds of millions of queries daily. That data trains Ernie. Ernie improves search results. Better search drives more users and more queries. The loop compounds. But the real value is not search revenue. It is the continuous training signal that improves the foundational models.
This repositioning also explains Baidu's capital allocation. A traditional search company would milk the business for cash and invest defensively. An AI company front-loading the stack invests aggressively in chip design, model training, and cloud infrastructure, accepting near-term margin pressure for long-term positioning.
The robotaxi ambition is the reveal. Baidu is not building robotaxis as a service business. It is building them as a proof-of-concept for the full stack. If Baidu can deploy Level 4 autonomy at scale in China, it validates the entire integrated approach. It also positions the company to compete with Tesla and Waymo globally.
He's framing—maximising token spend, prioritising safety—reads like a hyperscaler. Baidu is no longer a search company optimising for margins. It is a compute-intensive AI platform betting that sovereignty and integration will matter more than marginal efficiency.
