Baidu is emerging as a major AI chip player in China to fill the Nvidia gap (www.cnbc.com)

Baidu is fast positioning itself as a major domestic AI-chip supplier through its Kunlunxin unit, pitching chips and cloud compute as a vertically integrated stack for LLM training, inference, cloud and telecom workloads. The company already runs a mix of Kunlun and Nvidia hardware in its data centers to power ERNIE models, sells chips to third parties, and rents compute via its cloud. Baidu released a five‑year roadmap that begins with an M100 chip in 2026 and an M300 in 2027, and has begun winning orders (including from suppliers to China Mobile), prompting analysts at Deutsche Bank, JPMorgan and Macquarie to raise forecasts and valuations — JPMorgan projects Kunlun revenue could hit ¥8bn (~$1.1bn) in 2026 and Macquarie pegs the unit at roughly $28bn. The move matters because U.S. export controls have limited access to Nvidia’s top GPUs in China, and Beijing has discouraged use of even lower‑end H20 parts, creating a large, semi‑captive market for local suppliers. If Kunlun’s generations ship on schedule and perform competitively for LLM workloads, Baidu could both solve its supply constraints and become a strategic hardware provider across China’s hyperscalers. Key risks remain: China’s fabs (notably SMIC) lag TSMC in scale and process technology, so manufacturing and capacity bottlenecks could cap how quickly domestic chips replace imported GPUs.