Baidu used its second-quarter 2026 earnings call to define a new growth engine. AI Cloud Infrastructure revenue reached RMB 7.3 billion, and GPU Cloud grew faster still. Chinese buyers are switching to Kunlunxin chips as Washington restricts Nvidia sales and Beijing pushes domestic technology.
Kunlunxin is reportedly preparing a Hong Kong listing targeting a valuation of up to US$50 billion, while Baidu expects dual-primary status before the end of 2026.
Baidu’s second-quarter earnings call sounded like a demand story. Set against Chinese import controls, it is more exactly a supply story—one Beijing now controls. The chip that cannot clear customs is altering who sells to Chinese clouds.
That veto, not product superiority alone, is a key mechanism behind Baidu’s confidence in its Kunlunxin chip unit. A domestic accelerator business is being pulled into the space Nvidia can no longer reliably serve.
Baidu executives framed the shift as customer preference for domestic silicon driven by rising demand for inference, constrained supply, and the search for high-performance, reliable, and cost-efficient alternatives. The sharper reading is a captive market: Chinese cloud buyers face a restricted import path and a government that has already chosen the alternative. That is not a demand shock alone. It is a supply lock.
Baidu is building a capital structure around that lock. Its chip unit may soon trade on its own. For anyone tracking Nvidia’s China exposure, the call was less a financial update than a policy signal.
A market forced, not won
The U.S. Commerce Department’s Bureau of Industry and Security shifted in early 2026 to case-by-case approvals for Nvidia’s H200 and AMD’s MI325X. Those approvals carried conditions: security safeguards and third-party testing in the United States. According to Reuters reporting, Chinese customs has nonetheless blocked H200 shipments. The policy shift matters more than the licensing nuance.
For a procurement lead at a Chinese cloud provider, the real question is not CUDA versus Kunlunxin. It is whether the US-approved part clears customs.
IDC’s 2025 China AI accelerator data quantify the turn. Domestic GPU and AI chip vendors shipped around 1.65 million cards last year.
That volume translated into 41 percent of China’s AI accelerator server market. Nvidia still held the largest single share, but the shift is now state-assisted rather than purely competitive. Market share measures shipments, not software performance, and a domestic win on volume is not parity on training.
Baidu reported that GPU Cloud revenue grew 283% year-over-year, marking its fourth straight quarter of triple-digit growth. Dou Shen, executive vice president of Baidu’s AI Cloud Group, said demand for AI computing across training and inference remains strong while supply stays constrained.
The growth is real; the mechanism behind it is narrower than the call suggests.
Robin Li, Baidu’s co-founder and CEO, framed the company as AI-first, with AI cloud infrastructure and GPU cloud as core engines. CFO Haijian He said the planned dual-primary Hong Kong listing is intended to widen capital access, not merely satisfy investors.
Overall revenue rose just 4 percent in the quarter. That makes AI cloud the growth engine by default, not by choice.
Apollo Go crossed one million fully autonomous rides in the June quarter. That scale matters less today for revenue than for the regulatory data it generates. ERNIE, Baidu’s consumer AI assistant, grew daily active users by 83 percent year over year.
The chart below spells out the difference between the two kinds of silicon.
