Baidu now gets half its revenue from AI products and services. This shift shows that China’s business market is successfully cutting ties with Western chips and software. For years, Western analysts dismissed Baidu as a slow search giant. They thought it was struggling to copy Google’s ad model while chasing self-driving cars and AI.

The reality in China is quite different. People there see Baidu’s shift as a planned effort to build a sovereign computing stack. This plan aligns closely with state goals. Baidu integrates its Ernie large language models with its own Kunlun AI chips. This integration creates a complete system. As a result, US export controls on Nvidia chips matter much less to Chinese business customers.

This success represents China’s version of the classic mainframe lock-in. It applies this old concept to the new era of generative AI. IBM once ruled corporate America by controlling both the hardware and the operating system. Now, Baidu is securing its local market by offering a fully Chinese alternative. Its chips, its PaddlePaddle deep learning framework, and its applications are built to run together.

This complete system is very attractive to state-owned companies and local governments. These groups face strict rules to remove foreign technology from their work. Western observers look at raw model benchmarks to judge who leads in AI. Chinese firms care more about supply-chain safety and local rules. Baidu holds a clear home-court advantage in these areas.

Yet, the main risk to this business model is the local hardware supply chain. It may struggle with production volume and efficiency. Local factories must supply Kunlun chips at the right volume and cost. If they fail, Baidu cannot support its growing cloud business. High costs for hardware will quickly eat up the company’s profitable AI services revenue.

Relying on local factories makes Baidu weak. China’s chip sector faces production bottlenecks because of current limits on printing machines. To see if this AI revenue can last, watch three clear signs over the next two quarters. First, track the spending margins of Baidu’s AI Cloud division. See if hardware costs are rising faster than service revenue.

Second, track how often Baidu wins contracts from state firms like China Mobile or State Grid. It bids for these deals with its Kunlun-Ernie system. Finally, watch for any new deals between Baidu and local chipmaker SMIC. Look for details on the production volume of the next-generation Kunlun III chip.

This story appeared in AsiaAI.FYI Issue #71.