Baidu is accelerating its Kunlunxin chip program to reduce China reliance on foreign AI GPUs. The company plans a clear product roadmap and a stronger sales push to cloud and data center customers. This shift aims to close gaps created by restricted Nvidia exports and rising local demand for AI compute.
Baidu Kunlunxin Roadmap
Baidu has mapped a five-year plan for Kunlunxin chips. It is said that the company will launch M100 in 2026 and M300 in 2027. The roadmap sees Kunlunxin as the main chip stack of Baidu Cloud and its Ernie family of AI models. Today, Baidu has its models being operated with a combination of Kunlunxin silicon and Nvidia.
Baidu sells chips to third parties and rents computing from its cloud. The company positions Kunlunxin as a full-stack offering that covers chips, servers, and model hosting. Analysts and banks have started to raise estimates for the unit as orders from hyperscalers and carriers arrive. One investment bank expects chip revenue to expand sharply in 2026.
Domestic Chip Shortages
Major Chinese cloud and internet firms say chip supply is tight. Executives from large groups have flagged shortages of AI memory and accelerator parts. That has made domestic alternatives more attractive for companies that must scale model training quickly. Baidu stands to win in this context by offering chips that are available inside China and that avoid export limits.
The government is also encouraging local production and buying. That policy push increases the chance that Kunlunxin will pick up broader orders beyond Baidu internal use. Analysts note that Timely deliveries and competitive performance will be key. China still faces manufacturing gaps versus global leaders in advanced wafer nodes. That constraint will shape how fast any domestic chip can take market share from foreign suppliers.
Baidu’s plan does more than replace hardware. The company aims to sell an integrated stack to clouds and carriers. That approach bundles chips with software and data center capacity. It can shorten deployment cycles for customers who need ready-made AI systems. Several market participants now view Kunlunxin as one of the better-positioned domestic designs for large language model training and inference.
Baidu faces tests on timing and scale. Delivering chip generations on a strict calendar will determine whether the company moves from a strategic project to a major revenue stream. The next 12 to 24 months will show how well Kunlunxin meets customer demands and how the wider supply chain responds.