Baidu's Robin Li warns AI could become a bubble if profits don't follow
Baidu CEO Robin Li told the APEC CEO Summit that AI bubble fears could come true if heavy spending doesn't lift business profits. He pointed to Baidu's new Token Exchange.
Robin Li, chairman and CEO of Baidu, warned on Tuesday that fears of an AI bubble could become reality if heavy investment doesn't translate into higher profits for businesses. He made the remarks at the APEC CEO Summit in Shenzhen, saying AI spending hasn't yet produced significant revenue or profit gains for most companies.
There's no single deal here, despite how the story is sometimes framed. It's a warning from the head of one of China's biggest AI companies, paired with a product Baidu hopes will help fix the problem he describes.
What Li actually said
The key line: "If we fail to address this, the concern that AI might be a bubble could well become a reality." Li split the industry into four layers (chips, cloud infrastructure, large models and applications) and said competition is currently fiercest at the foundation level, while value ultimately has to be captured at the application layer.
He said the spread of AI applications has fallen short of expectations despite heavy infrastructure spending, rapid model development and rising token use. His prescription is organisational: companies need to rethink how they structure work to turn AI capabilities into business gains. One report says he called the situation "urgent".
He isn't calling the technology a bust. Li said he remains optimistic that AI is an "inevitable trend".
The Token Exchange angle: who pays, who gains
Li also pointed to Baidu's Token Exchange. Baidu says it connects AI token providers directly with enterprise and individual customers to help ease computing shortages. Customers can compare model services side by side with clear pricing and performance data, and providers get a channel to more buyers.
So customers pay for tokens, providers get paid, and Baidu runs the marketplace. The platform verifies model capabilities and rates providers through regular testing and real-time monitoring. On the earnings call, management said it expects the exchange to help put underused compute to work. Baidu hasn't spelled out its own cut in anything I've read, so I won't guess at it.
The usage numbers come from Baidu itself. After one month, the exchange had onboarded more than 100 model resources and was handling an average of over 50 million calls and 700 billion tokens a day. That's a lot of volume, but token counts say little about whether the buyers are making money.
Baidu's own numbers
Li's company sits on both sides of this argument. Baidu's third-quarter results showed 94% year-over-year growth in AI cloud infrastructure revenue. Yet Quartz's headline notes the company posted its sixth straight revenue drop even as the AI cloud business grew. Li acknowledged headwinds in online marketing while saying the AI strategy had produced tangible results.
That mix is probably why the message lands. Infrastructure revenue is growing fast, while the wider business still depends on whether AI pays off for customers.
The stakes for the AI market
Li's comments came just before Xi Jinping and Donald Trump were due to hold talks during APEC, where AI risks are expected to be on the agenda. One summary also links his remarks to Kristalina Georgieva's earlier warning about shaky trust in AI investments.
Honestly, a CEO who sells AI infrastructure saying most customers aren't seeing profit gains is worth noticing. He has every reason to stay upbeat, and he's still flagging the gap. I'd read it as a pitch as well as a warning: the answer to the bubble question, in Li's telling, includes Baidu's marketplace and Baidu's applications.
It also matches a broader pattern. Li said at Baidu World that more than one new model was released almost every day in the first nine months of the year. Supply of models is clearly not the bottleneck. Proof that businesses earn more from using them is, and I think that's the number investors will start asking for.
What to watch next
Two things matter. First, whether the Xi and Trump talks produce anything concrete on AI risk. Second, whether Baidu or its rivals can publish customer results that show actual revenue or profit gains, rather than token volumes and API calls.
If you're deciding where to put AI budget, I'd wait for that kind of evidence before committing big. Li's warning is a sensible filter: ask what your AI spending is meant to change on the profit line, and by when.