The real money lies in solving concrete problems, not chasing raw power
In a quiet but consequential turn, two veterans of China's AI establishment have stepped away from the race to build ever-more-powerful general intelligence and are instead asking a more grounded question: what specific burdens can machines lift from specific human lives? Yoolee AI and InfiX.ai, positioning themselves against US startup Thinking Machines Lab, are wagering that the future of artificial intelligence is not found at the frontier but in the unglamorous, lucrative work of solving concrete problems in healthcare, travel, and enterprise — where the cost of human labor has grown too heavy to bear. Their pivot reflects a maturing reckoning within the global AI industry: that competitive advantage may belong not to those who build the largest minds, but to those who build the most useful ones.
- The frontier AI race — defined by ever-larger models and astronomical compute costs — is quietly losing its most pragmatic participants, as seasoned Chinese researchers conclude the real opportunity lies in narrower, more deployable solutions.
- Yoolee AI and InfiX.ai are entering direct competition with Thinking Machines Lab, the US startup founded by former OpenAI executive Mira Murati, raising the stakes in a vertical AI market that is rapidly attracting serious capital.
- The pressure is not only competitive: Chinese AI companies face persistent concerns around data privacy and regulatory scrutiny, making self-contained, enterprise-grade AI agents a strategically safer bet than cloud-dependent general-purpose systems.
- Yoolee's model — blending task-focused AI agents with Palantir-style enterprise data infrastructure — signals an attempt to capture both the intelligence layer and the deployment layer of business AI adoption.
- With venture backing secured and a clear market thesis in place, the central uncertainty is speed: US competitors already hold footholds in vertical markets, and the window for Chinese startups to establish themselves may be narrowing.
Two former leaders of Chinese AI research labs have made a deliberate retreat from the frontier — the high-stakes competition to build ever-larger, ever-more-capable general AI systems — and are instead building something more modest and, they believe, more valuable: machines that know how to do specific jobs well.
Yoolee AI, led by Zhang, a former chief operating officer at Beijing-based Zhipu AI, and InfiX.ai are both positioning themselves as Chinese counterparts to Thinking Machines Lab, the US startup founded by Mira Murati after her departure from OpenAI. The argument their founders make is direct: the real money is not in raw computational power but in AI agents that can manage patient intake in a hospital, plan and book travel itineraries, or handle the operational workflows that have made human labor prohibitively expensive across entire industries.
Yoolee's pitch is particularly layered. Zhang describes the company as building self-evolving agents — systems that improve through use — while also providing the enterprise data infrastructure needed to deploy them inside existing organizations. He draws a deliberate parallel: the task-focused intelligence of Thinking Machines Lab, combined with the institutional-scale data management that Palantir offers governments and corporations. Lanchi Ventures and other backers have found the thesis credible enough to fund.
The shift carries strategic logic beyond market opportunity. Specialized AI models are smaller and cheaper to run than general-purpose ones, easing the cost pressures that have made frontier AI so punishing. And by keeping data within a company's own systems rather than routing it through external cloud providers, these startups sidestep the privacy and regulatory sensitivities that have long complicated Chinese AI's relationship with enterprise clients.
What remains unresolved is timing. US competitors have already begun establishing themselves in vertical markets, and the question facing Yoolee, InfiX.ai, and others is whether Chinese AI's deep technical talent and access to venture capital can translate into market position quickly enough — and whether China's own regulatory environment will give them the room to find out.
Two former leaders of Chinese AI research labs have stepped back from the race to build the most powerful general-purpose artificial intelligence systems and are instead betting on a narrower, more practical path: teaching machines to handle specific jobs within specific industries.
Yoolee AI and InfiX.ai, both launched by veterans of China's AI establishment, are positioning themselves as rivals to Thinking Machines Lab, a US startup founded by Mira Murati, who previously worked at OpenAI. The Chinese founders argue that the real money and impact lie not in chasing raw computational power but in building AI agents tailored to solve concrete problems in healthcare, travel, finance, and other sectors where human labor has become prohibitively expensive.
Zhang, who leads Yoolee AI, describes the company's mission as creating self-evolving AI agents—software systems that can learn and improve on their own as they work. A healthcare-focused agent might manage patient intake, appointment scheduling, and basic diagnostic triage. A travel agent could handle itinerary planning, price comparison, and booking logistics. The pitch is straightforward: these tools could generate new revenue streams for businesses while cutting operational costs. Yoolee raised backing from venture firms including Lanchi Ventures, signaling investor confidence in the vertical AI model.
Zhang's own trajectory reflects a broader shift in how Chinese technologists are thinking about artificial intelligence. He spent 2023 as chief operating officer at Zhipu, a Beijing-based AI lab, before departing to start Yoolee. In describing his new venture, he draws an interesting parallel: it combines the specialized, task-focused approach of Thinking Machines Lab with the enterprise data management infrastructure that Palantir, the US software giant, provides to governments and large corporations. In other words, not just smart agents, but the systems to deploy them at scale within existing organizations.
This pivot away from frontier AI—the high-stakes competition to build ever-larger language models and reasoning systems—suggests that some of China's most talented AI researchers have concluded the frontier is crowded and expensive, and the real competitive advantage lies elsewhere. Building industry-specific solutions also addresses two concerns that have dogged Chinese AI companies: cost and privacy. Specialized models can be smaller and more efficient than general-purpose ones, reducing computational overhead. And keeping data within a company's own systems, rather than sending it to a cloud provider or third party, addresses regulatory and competitive sensitivities.
The timing is significant. US startups like Thinking Machines Lab have already begun attracting attention and capital by focusing on vertical applications rather than chasing OpenAI or Anthropic at their own game. Chinese competitors, with deep expertise in machine learning and access to substantial venture funding, are now making the same calculation. The question is whether they can move fast enough to establish themselves in markets where US competitors already have a foothold—and whether the regulatory environment in China will allow them to operate freely as they scale.
Notable Quotes
Zhang described Yoolee as combining Thinking Machines Lab's specialized approach with Palantir-style enterprise data management infrastructure— Zhang, founder of Yoolee AI
The Hearth Conversation Another angle on the story
Why would experienced AI researchers step away from frontier models to build industry-specific tools? That seems like a step backward.
It's actually the opposite. Frontier AI is a winner-take-most game—you need billions in compute and talent to compete with OpenAI or Anthropic. But industry-specific agents solve real problems that companies will pay for today. The margins are better, the path to revenue is clearer.
So this is about pragmatism, not capability?
Both. A healthcare AI agent doesn't need to be a general reasoning system. It needs to be good at one thing, and it needs to be cheaper and faster than hiring people. That's actually harder to do well than it sounds.
What's the privacy angle?
If you're a bank or hospital, you don't want your data going to a third-party cloud service. These startups are offering to build and run AI systems inside your own infrastructure. That's a huge selling point in regulated industries.
And China specifically—does this give them an advantage?
They have the engineering talent and venture capital. But they also face regulatory constraints that US companies don't. Building specialized tools for specific industries might actually be easier to navigate than trying to export general-purpose AI systems.
So we're looking at a fragmented AI market, not a consolidated one?
Exactly. Instead of one or two dominant platforms, you'll have dozens of specialized players, each owning a vertical. That's a very different competitive landscape.