No Time for Domestic AI to Rest on Laurels

Continuous investment in artificial intelligence infrastructure is essential, even as emerging models like DeepSeek's open-source R1 challenge existing paradigms of efficiency. Robin Li, the CEO of Baidu Inc., highlighted the necessity of significant investment in areas such as chips, data centers, and cloud services at the World Governments Summit 2025 held in Dubai.

Li emphasized that given the fast-evolving nature of technology, companies cannot afford to halt their investment efforts. "When technology continues to advance rapidly, you must keep investing to remain at the forefront of this innovation era," he stated.

He elaborated on the nature of AI development, mentioning that exploring various methods may lead to breakthroughs that make model training cheaper. However, initial investments can be substantial. According to him, it could take billions to identify an efficient approach, one that ultimately reduces model training costs, such as down to $6 million.

DeepSeek's cost-efficient model has caused a stir, performing comparably to US-based OpenAI's offerings but at a fraction of the cost and required computing power. This development raises pertinent questions about the essentiality of large-scale AI infrastructure.

Innovation thrives best in environments that encourage experimentation, as noted by Li. He stated, "Innovation cannot be scheduled; you never know when it will occur. What’s crucial is to create a nurturing atmosphere for innovation to flourish."

Baidu has been proactive in AI advancements, launching its large language model (LLM), Ernie Bot, in March 2023, shortly after the unveiling of OpenAI's ChatGPT in November 2022. Li remains hopeful about AI's future, asserting that LLMs have the potential to add substantial value in various applications. He noted that the costs associated with using foundational models can decrease by over 90% within a year.

According to Li, there is a direct relationship between reducing costs and increasing productivity. If operational expenses drop significantly, productivity generally follows suit, a principle he claims is fundamental to innovation.

In the enterprise sector, numerous businesses have leveraged LLMs to enhance operational efficiency, achieving results at just one-tenth of the prior costs.

Experts, such as Lu Yanxia from IDC China, agree that ongoing investments and technological advancements by Chinese tech firms will enhance the accessibility of AI applications and open up new business avenues for domestic AI services, cloud computing, and chip manufacturers. Open-source models are particularly beneficial, aiding enterprises and developers in speeding up AI innovation.

Despite US restrictions on exporting advanced chips, analysts like those from CITIC Securities assert that domestic AI models continue making significant advancements due to their cost-effectiveness, high performance, and open-source advantages, leading to broader applications across diverse fields.

When examining AI deployment opportunities, Li pointed towards autonomous vehicles, especially robotaxis, asserting they could transform transportation by dramatically improving safety. "We have demonstrated that robotaxis are considerably safer than human-operated vehicles. Today's robotaxis are at least ten times safer," Li remarked. Moreover, he highlighted that Baidu's insurance claim rates are remarkably lower compared to traditional taxis.

Baidu has heavily invested in autonomous driving technology. Its robotaxi service, Apollo Go, currently operates in 11 cities in China, and fully autonomous robotaxis manage operations in select areas of cities like Beijing, Chongqing, Wuhan, and Shenzhen. Li remarked that Baidu is ready to expand robotaxi operations wherever regulations permit, reinforcing the narrative of continuous growth and development in the AI field.

AI, Investment, Innovation