Baidu’s Robin Li and Xiaomi’s Lei Jun Warn AI Returns Lag Spending

At the APEC CEO Summit in Shenzhen, Li said AI could become a bubble if most adopters see no profit gain, a day after Baidu’s sixth straight revenue decline.

WHY IT MATTERS

Two of China's largest AI spenders now tie infrastructure budgets to profits from AI use, putting Baidu's application revenue, flat at RMB2.6 billion last quarter, under closer scrutiny.

Baidu founder Robin Li told the APEC CEO Summit in Shenzhen on November 17 that artificial intelligence will turn out to be a bubble if most companies that adopt it fail to grow their profits. Caixin, which reported his remarks from the venue, said Li described progress in 2026 as "urgent" because the spread of AI applications has fallen short of expectations while competition in models and infrastructure intensifies. Xiaomi founder Lei Jun, speaking on a separate panel, made a similar point about the distance between spending and returns.

The warnings came from two executives whose companies spend heavily on AI. The South China Morning Post, which is owned by Alibaba, noted that Li spoke a day after Baidu reported its sixth consecutive quarterly revenue decline. Baidu’s third-quarter results showed AI cloud revenue up 94% while online marketing revenue fell 22%, and its Hong Kong shares fell to their lowest level in more than five years on November 17.

Li’s argument puts the burden of proof on the companies buying AI rather than on the labs building it. Neither he nor Lei gave a figure for how far returns trail spending, and neither said their own companies would cut investment.

Li says spending doubles yearly while adoption lags

Robin Li said spending on AI has risen fast enough over the past three years that annual outlays may have doubled or more each year, according to a Sichuan Daily report carried by The Paper. He listed the pressures at the infrastructure end: models are replaced quickly, the industry is pouring money into computing, rents for cloud infrastructure have recently risen, and token consumption is climbing because many companies have bought usage plans for their staff. Most companies using AI, he said, have not seen a clear increase in revenue or profit.

"If we fail to address this, the common fear that AI might be a bubble could well become a reality," Li said, as quoted by the SCMP. He called finding high-value applications the most pressing task. Baidu’s own results offer a partial test. AI accounted for 53% of revenue from Baidu’s general business in the third quarter, up from 39% a year earlier.

Li also challenged the cost of the US-China contest to build ever more advanced models, according to the SCMP, and compared AI with electricity, whose economic gains arrived decades later once factories were reorganized around it. Companies need "an organisational structure that aligns with AI’s capabilities to derive real value," he said. Caixin’s headline said Baidu had made "radical" internal changes along these lines, though the report gave no details. TechNode noted that Li had raised the same theme at Baidu World the previous week.

Lei Jun sees few companies closing the commercial loop

Lei Jun was blunter about the industry’s economics. "The investment in AI has been extremely huge, yet the scenarios where the technology can actually be implemented are very limited," he said, according to the SCMP, adding that very few companies had found a way to turn the technology into profit. The Paper reported him saying many companies are troubled and "very anxious" as a result.

Lei said four years of rapid model development had yet to produce a revolutionary, benchmark-setting product combining AI with phones or cars, though he expected great products in both within a few years. Xiaomi’s own bet is on AI moving into physical machines. Lei said Xiaomi has applied AI across its factory processes, has built a factory that runs without workers and is exploring an industrial agent that could run production, find problems and improve itself. He also predicted that humanoid robots could bring far-reaching changes to manufacturing, according to the SCMP.

Baidu is betting on agents and its own chips

Li named the uses he thinks will pay first. Domain-specific AI agents are more likely to produce measurable benefits, he said, citing Baidu’s work with the Port of Qingdao, where agents work out the most efficient way to load containers. He also said computing power remains in short supply across the industry and at Baidu, which plans to launch a 1,024-chip Tianchi supernode powered by Kunlunxin processors next year, according to the SCMP.

Baidu’s position in models is under pressure. The SCMP’s assessment is that DeepSeek, Moonshot AI, Tencent and Alibaba have overtaken Baidu’s Ernie foundation models, even after last week’s Ernie 5.5 Flash Preview release. When CATL’s Robin Zeng asked Li whether current models are good enough, Li replied that AI is good at solving problems but that posing and checking the right questions matters more, The Paper reported.

Other speakers showed where returns are already being booked. TCL said five years of AI use in display manufacturing had produced about RMB2.5 billion in benefits over the past three years. Baidu’s next quarterly report will show whether its AI applications revenue, which was roughly flat in the third quarter at RMB2.6 billion, starts to move as Li says the whole industry must.