A distinct strategic divergence is emerging within the Chinese artificial intelligence ecosystem. While heavily funded giants like DeepSeek and Baidu continue to pour billions into the race for artificial general intelligence (AGI), a new wave of startups founded by former lab leaders is explicitly stepping back from the frontier model competition. Instead, these companies are pivoting toward industry-specific, privacy-focused AI solutions designed to deliver immediate, tangible business value.
This shift reflects a growing recognition that pursuing absolute general intelligence is not only prohibitively expensive but may also be unnecessary for the vast majority of enterprise applications. Startups like Yoolee AI and InfiX.ai are pioneering this pragmatic approach, aiming to solve the “last mile” of generative AI deployment for businesses that require specialized capabilities, stringent data security, and cost-effective operation.
Yoolee AI: The Palantir of Industry Agents
Yoolee AI, founded in 2025 by Zhang Fan, the former chief operating officer of Zhipu AI, exemplifies this new strategic direction. Backed by Lanchi Ventures, Yoolee is focused on building self-evolving AI agents tailored for specific industries, such as personal healthcare specialists and travel planners.
In an interview with the South China Morning Post on the sidelines of the LEAP East exhibition in Hong Kong, Zhang articulated the rationale behind the pivot. “All models are relentlessly chasing the ceiling of general intelligence … but in many real-world scenarios, we don’t necessarily need an absolute all-rounder,” he stated.
Zhang observed that while current frontier models are “extremely powerful,” very few major enterprises have actually seen AI generate tangible business value. Yoolee aims to bridge this gap by deploying agents capable of performing complex, industry-specific tasks that are currently too costly for human workers, thereby driving new revenue streams for its clients.
Zhang described Yoolee as a combination of Thinking Machines Lab—the highly watched US startup founded by former OpenAI CTO Mira Murati—and Palantir, the US data software giant known for helping large enterprises and governments manage complex, secure AI systems.
InfiX.ai: Conquering the Last Mile of GenAI
A similar philosophy drives InfiX.ai, a Singapore-headquartered startup founded by Professor Hongxia Yang, a prominent AI researcher at Hong Kong Polytechnic University (PolyU) and former leader at the Samsung China Research Laboratory. InfiX.ai is focused on what Yang calls the “last mile of GenAI”, adapting foundation models for enterprise use without compromising proprietary data.
InfiX.ai’s core technology revolves around low-bit training and model fusion, an approach Yang terms “Co-GenAI” (collaborative generative AI). This technology is designed to lower barriers to developing generative AI, enabling organizations to train and adapt models using their own data and expertise without relying on massive, centralized compute clusters.
“If you rely on centralization, no collaboration happens; people just compete with each other,” Yang noted in a recent company insight. Her vision is a future of decentralized AI, where “consumers as well as enterprises, particularly small and medium-sized enterprises, to have their own domain-specific models.”
The company’s flagship product, InfiCube, is an integrated large-scale training and inference appliance designed to enhance data privacy through localized ownership and optimize ROI through smarter computing scheduling. The technology recently won the BEYOND Best of Innovation Award 2026 and was recognized in the Top 30 for the WAIC 2025 SAIL Award.
The Economics of Pragmatism
The pivot toward industry-specific AI is driven by harsh economic realities. Training frontier models requires massive capital investment and access to tens of thousands of advanced GPUs, resources that are increasingly scarce due to US export controls. As DeepSeek targets a $71 billion valuation to fund its infrastructure needs, smaller startups are realizing they cannot compete on raw compute power.
By focusing on domain-specific models and efficient training techniques like low-bit quantization, companies like Yoolee and InfiX.ai can deliver high-performance solutions at a fraction of the cost of frontier models. This pragmatic approach is highly attractive to Chinese enterprises, which are eager to adopt AI but are highly sensitive to operational costs and data security risks.
Furthermore, the focus on on-device and localized AI aligns with broader hardware trends. InfiX.ai’s recent partnership with Anker Innovations to establish a joint laboratory at PolyU underscores the growing importance of deploying efficient AI models directly on smart home products and edge devices, thereby reducing reliance on cloud infrastructure.
A Bifurcated Ecosystem
The emergence of startups like Yoolee and InfiX.ai suggests that the Chinese AI ecosystem is bifurcating. On one side are the heavily capitalized giants pursuing AGI and building massive, centralized foundation models. On the other side is a growing cohort of agile, pragmatic startups focused on decentralized, industry-specific applications.
This bifurcation may ultimately prove to be a significant strength for the Chinese AI industry. While the giants push the boundaries of raw capability, specialized startups are doing the difficult work of integrating AI into the fabric of the traditional economy, ensuring that the technology actually delivers on its promise to drive productivity and business value.