Baidu established the Institute of Deep Learning in 2013. In 2014, the company appointed Andrew Ng as chief scientist to lead Baidu Research. This research umbrella includes the Silicon Valley AI Lab, the Beijing Deep Learning Lab, and the Beijing Big Data Lab. Ng previously founded and led the Google Deep Learning team. His work at Google involved building massive artificial neural networks. One specific project used a neural network with 1 billion parameters to detect cats from unlabeled YouTube videos. At Baidu, Ng managed a team of 1,300 AI researchers. The Beijing Deep Learning Lab, which was formerly the Institute of Deep Learning, focuses on image recognition, voice recognition, natural language processing, and machine translation. Ng also co-founded Coursera, which reached 9 million students. He also taught an online Machine Learning class to over 100,000 students at Stanford University. Ng was named to the Time 100 list and Fortune’s "40 under 40" list. Baidu’s search platform saw media spend growth of 4% year-over-year in the first quarter of 2014. This search engine holds roughly 24% of all clicks and 7% of all impressions globally. In that same quarter, Baidu held about 8% of the global search media spend.
The Silicon Valley migration
The tech ecosystems in Silicon Valley and Beijing operate as separate universes. Researchers migrate between these worlds as commercialization pressures rise. Many senior researchers leave established firms like OpenAI and DeepMind to join startups like Anthropic. Anthropic maintains a safety-first ethos that attracts researchers who find big tech too corporate. At OpenAI, researcher Zoe Hitzig resigned because the company planned to introduce advertising on ChatGPT. OpenAI also disbanded its Mission Alignment team 16 months after its establishment. Jerry Tworek left OpenAI in January to pursue research outside the company. The decision of researchers to leave large organizations often stems from a desire for creative freedom and a focus on safety that they feel is lost when research becomes secondary to product timelines and investor demands. You already know how fast these shifts happen. When a lab loses enough mass, the center of gravity shifts to where the compute and authorship align. Anthropic completed a 30-billion-dollar funding round and has a valuation of 380 billion dollars. OpenAI is currently in talks to raise 100 billion dollars. Elon Musk’s xAI also saw multiple co-founders and engineers resign during organizational restructuring.
Chinese models challenge US dominance
Baidu’s ERNIE 4.5 outperforms OpenAI’s GPT-4.5 in multiple benchmarks. This model costs only 1% of the price of GPT-4.5. ERNIE 4.5 demonstrates collaborative optimization through joint modeling of multiple modalities and exhibits improvements in understanding, generation, reasoning, memory, and hallucination prevention. The ERNIE X1 model is a multimodal deep-thinking reasoning model that matches the performance of DeepSeek R1. ERNIE X1 excels in Chinese knowledge Q&A, literary creation, and logical reasoning. It uses the Progressive Reinforcement Learning Method and an End-to-End Training Approach Integrating Chains of Thought and Action. The model also uses a Unified Multi-Faceted Reward System and possesses capabilities like planning, reflection, evolution, and TreeMind mapping. As of December 2025, China had 602 million generative AI users. Baidu AI search has 365 million monthly active users. Baidu’s ERNIE Bot rivals OpenAI’s ChatGPT. Baidu also launched a $145 million venture capital fund named Baidu Ventures. This fund focuses on companies with products like speech recognition and natural language processing. Alibaba has also released an open-source AI reasoning model called QWQ-32B. This model outperforms DeepSeek R1 in coding. Baidu’s search product introduces AI Mate to help users with their searches. For brands, Baidu uses the DuXingxuan platform to manage representation within AI answer systems. This platform uses more than 12 million Baijiahao authors to provide content. The system manages the consumer journey through four stages: cognitive, demand, contrast, and decision. It helps brands address questions about medical treatments, hotels, and skincare products. Will the massive US investment in compute eventually create an insurmountable gap?
