Economic Watch: Chinese AI firms start to crack profit code in niche markets

SHANGHAI,平阳消防新闻网 林学有 March 28 (Xinhua) -- While Silicon Valley chases artificial general intelligence with billion-dollar training runs, a burgeoning crop of Chinese AI firms has been achieving what the entire industry is striving for: sustainable profitability in specialized markets.

Thanks to China's industrial scale and market volume, these notoriously R&D-heavy tech ventures are now cashing in on their early commercial viability.

A financial report this week brought an encouraging glimpse into the momentum. XtalPi, an AI drug discovery company listed in Hong Kong, turned a profit of 134.6 million yuan (approximately 19.5 million U.S. dollars) in 2025, following a loss of 1.5 billion yuan the year before. This turnaround occurred even as the tech firm increased its R&D expenditure by approximately 36 percent.

Such a success serves as a microcosm of China's broader strategy in the global tech race: rapidly integrating AI with its robust industrial base to capture premium advantages in specialized sectors.

XtalPi, leveraging AI to facilitate drug discovery for global biopharmaceutical firms, made multiple commercial progress last year, including advancing targeted cancer therapies and pediatric disease treatments into clinical trials.

"The number of our revenue-generating clients increased by 62 percent year on year," XtalPi's CFO Zhou Feiran told Xinhua, explaining the turnaround to profitability.

In 2025, the company advanced five drug pipelines into clinical stages, and deployed AI agents capable of independently running over 10,000 experiments per week.

Wen Shuhao, chairman of XtalPi said: "Our AI-Robot-Data-R&D flywheel has demonstrated powerful value across multiple fields, including biopharmaceuticals and new materials."

INDUSTRIAL BASE

XtalPi's expansion in the "AI + Science" business is unfolding against the backdrop of China's rapidly growing biopharmaceutical sector. Official data show that China's outbound licensing deals for innovative drugs exceeded 130 billion U.S. dollars in total value in 2025, setting a record high.

This week, the U.S.-China Economic and Security Review Commission released a research paper titled "Two Loops: How China's Open AI Strategy Reinforces Its Industrial Dominance" that corroborates the dynamic.

The research argues that China's open-source AI strategy and its manufacturing base are forming two "mutually reinforcing" feedback loops, with its industrial base generating "interlocking innovation flywheels" across adjacent sectors.

"Beijing's AI+ Initiative, its robotics industrial policy, and its institutional infrastructure for converting deployment data into a national asset all reflect a strategy organized around deployed AI," according to the report.

China's manufacturing sector has remained the world's largest for 15 consecutive years, and its 15th Five-Year Plan (2026-2030) underscores the preservation of a robust manufacturing share in the economy and the establishment of a modern industrial system anchored in advanced manufacturing.

By targeting opportunities in "AI plus manufacturing," Shanghai-based AI unicorn Blacklake achieved profitability in late 2024 and has maintained rapid profit growth since then.

Blacklake co-founder Zhou Yuxiang recalled visiting a major factory in central China, when asked why he chose this business track. "While frontline workers used WeChat in their personal lives fluently, the production floor still relied on paper for record-keeping, and issues were often just shouted across the room."

This experience led Zhou to systematically gather production and quality data, then apply AI to enable precise, data-driven decisions.

Blacklake now serves 30,000 factories, helping small and medium-sized manufacturers overcome challenges in digital transformation. "These SMEs are the backbone of China's exports," said Zhou.

"By leveraging information technology to innovate their supply chains, we are striving to strengthen their risk resistance in an increasingly volatile global market," he added.

HUGE MARKET

China's vast domestic market provides fertile ground for AI adoption. Tech giants like Alibaba have begun integrating agentic AI, linking it with their existing digital infrastructure -- spanning payments, logistics, and cloud services -- to convert algorithms into revenue.

Unitree, the intelligent robotics maker that captivated audiences with its dance performance during China's Spring Festival gala last month, has also reported impressive financial results recently. Its prospectus for an upcoming IPO revealed a net profit exceeding 600 million yuan in 2025, a six-fold increase from the previous year.

The shift in revenue share highlights the trend: Unitree quadruped robots' income from research and education fell from 69 percent in 2022 to 32 percent in the first nine months of 2025, while consumer and industrial application shares rose to 42 percent and 26 percent, respectively, in the latter period.

Also, its quadruped robots are increasingly adopted for smart inspection and fire emergency response and are gradually expanding into areas such as survey detection and smart manufacturing, according to Unitree.

Chinese local governments have shown a willingness to ease regulations to accelerate AI adoption in consumer-facing sectors such as autonomous driving. Early this month, Pony.ai announced that its new-generation Robotaxi service achieved positive monthly per-vehicle profitability in Shenzhen, south China, in February, following a similar milestone reached in nearby Guangzhou last November.

As of February 16, this AI firm's paid Robotaxi orders in Shenzhen had already exceeded the company's total for all of 2025. "This demonstrates that driverless ride-hailing is not only technologically viable but has also reached economic sustainability for scaled operations," said Peng Jun, Pony.ai's founder and CEO.

China's strategy prioritizes deploying embodied AI in sectors "where specialized, real-world data from widespread use may compound into advantages that proprietary U.S. models cannot easily replicate, even if they maintain technical superiority on benchmarks," according to the afore-mentioned U.S. paper. ■