July 14, 2026, (Inside AI) — Two Chinese AI startups are pivoting away from the costly frontier model race to directly challenge Thinking Machines Lab, the US firm founded by former OpenAI executive Mira Murati. Yoolee AI and InfiX.ai are betting on industry-specific agents, lower costs, and stronger privacy to carve out a defensible niche.
The move signals a strategic split in China’s AI sector. While giants like Baidu and Zhipu pour billions into scaling large language models, these newcomers argue that value lies in vertical applications—not chasing ever-larger parameter counts.
Yoolee AI, backed by Lanchi Ventures, was founded last year by Zhang, who previously served as chief operating officer at Beijing-based Zhipu. He describes the company as a hybrid of Thinking Machines Lab and Palantir—combining advanced AI research with enterprise data integration.
Zhang told reporters that Yoolee builds self-evolving AI agents for tasks currently too expensive for human labor. Examples include personal healthcare specialists and travel planners, which he says can unlock new revenue streams for businesses in those sectors.
InfiX.ai, founded by another ex-Chinese lab leader, is taking a similar path. The company focuses on privacy-preserving AI for finance and legal sectors—industries where data sensitivity often blocks adoption of foreign models.
Both founders argue that the frontier race has become unsustainable. Training costs for models like GPT-5 exceed $1 billion, while returns remain uncertain. By contrast, fine-tuning smaller models on proprietary data can deliver 90% of the performance at a fraction of the cost.
A Deliberate Departure from the Scaling Orthodoxy #
This shift mirrors a broader debate in AI. Last year, a paper from Stanford’s Center for Research on Foundation Models found that vertical fine-tuning can match general-purpose models on domain-specific benchmarks. Chinese labs appear to be internalizing that lesson faster than their US counterparts.
Zhang’s comparison to Palantir is telling. Palantir’s market cap has surged past $200 billion by solving messy enterprise data problems, not by building the largest models. Yoolee aims to replicate that playbook with a Chinese twist—offering agents that adapt to local regulatory and linguistic nuances.
Yet the path is not without risk. Thinking Machines Lab has raised over $2 billion and is aggressively courting Asian enterprises. Murati’s team recently opened a Tokyo office and is reportedly developing lightweight models optimized for on-device deployment.
Privacy could be the deciding factor. Chinese enterprises face strict data localization laws under the Personal Information Protection Law. A homegrown solution that keeps data in-country may win trust faster than a US rival, even one with superior technology.
Industry analysts remain cautious. “The Chinese AI market is crowded with well-funded players,” said Li Wei, senior analyst at TechAsia Research. “Yoolee and InfiX need to prove they can deliver measurable ROI before clients switch from incumbents like Baidu AI Cloud.”
Both startups are still in early stages. Yoolee has pilot projects with two unnamed healthcare providers, while InfiX.ai claims a contract with a top-tier Chinese law firm. Revenue figures have not been disclosed.
The timing may be opportune. Global enterprise AI spending is projected to hit $300 billion by 2027, according to IDC. If Chinese labs can capture even a slice of that market with specialized agents, they could build sustainable businesses without winning the frontier war.
Zhang remains bullish. “We don’t need to beat GPT-6. We need to solve real problems that generate cash,” he said. That pragmatism, if executed well, could redefine what leadership in AI looks like.