A Nanjing startup is betting that the next AI infrastructure challenge will be coordinating models rather than merely serving them. SinapisAI, also called Nanjie Intelligent, has raised an angel round of nearly 100 million yuan, according to a September 10 report by the Sci-Tech Innovation Board Daily carried by NetEase. The company says it is working on underlying large-model algorithms, model-capability characterization, massive AI collaboration, and a general AI technology base. Its funding announcement is an early signal of how Chinese investors are beginning to finance the systems intended to connect many specialized models and agents.
Dingxin Capital and Nanjing Innovation Investment Group co-invested in the reported round. The company was founded in 2026 and is based in Nanjing. Its age makes the financing notable, but it also limits what can be concluded. The cited coverage provides no valuation, customer list, commercial deployment, revenue, or named pilot. The company is presenting an architectural ambition, not reporting an established platform with a public usage record.
The name SinapisAI is more than a branding detail in this case. The company’s concept is to make distributed AI capabilities understandable and composable on demand, according to the cited reports. That places it in the emerging layer between model builders and end users: the orchestration systems that determine what an agent can call, how it shares context, and how several models divide a task. The result could be useful for enterprise workflows, robotics, scientific tools, or public systems. The announcement does not state which market SinapisAI will target first.
A Multi-Agent Infrastructure Bet Built Around Learnware
SinapisAI’s technical narrative draws on “Learnware,” a concept associated with Zhou Zhihua, an academician of the Chinese Academy of Sciences and a professor at Nanjing University. The term describes reusable learning components that can be understood, selected, and combined for a task. The company’s cited materials frame that idea as a route toward letting large numbers of AI agents collaborate without every capability being built again from the beginning.
This is a demanding proposition. A multi-agent system needs more than a collection of model APIs. It has to identify what a model can do, decide when to use it, manage context, coordinate tools, and handle failure. If an agent calls an incorrect tool or combines incompatible outputs, the overall system can become less reliable even when the individual models are capable. SinapisAI’s financing is meant to support research and development, product construction, and market expansion, but the company has not disclosed technical evaluations that show how its approach performs in practice.
The problem is attracting attention across China’s AI ecosystem. EastFrontier recently examined TokenRythm’s NeoHorse-1 for agentic post-training, which approaches agent performance through model training. SinapisAI is describing a different layer: the infrastructure that could allow multiple capabilities to be discovered and coordinated. The two approaches are complementary in principle, but neither announcement proves that the full agent stack has been solved.
Nanjing Roots Link Research Talent to Commercial Infrastructure
The company’s reported leadership includes chief executive Sun Tengzhong and chief technology officer Liu Jiandong, both described in the cited coverage as alumni of Nanjing University’s LAMDA Lab. That academic connection helps explain the emphasis on reusable learning components and model characterization. It does not itself guarantee a commercial advantage. A startup still needs to turn research concepts into tools that developers can integrate, test, operate, and trust.
Nanjing has become a meaningful location for China’s AI research and startup activity, with university labs, public capital, and local technology programs creating a base for early companies. Nanjing Innovation Investment Group’s participation gives SinapisAI a local institutional investor alongside Dingxin Capital. The announcement does not describe governance terms or how the funds will be allocated between research, computing, hiring, and sales. Those details matter because AI-infrastructure startups can spend heavily before they find a repeatable enterprise customer.
The company is entering a market where the hardware and software layers are both moving quickly. China has invested in data centers, domestic chips, and agent models, while enterprises increasingly want software that can use those components without building every integration in-house. EastFrontier recently covered China’s 2030 road map for AI infrastructure, which highlights the policy backdrop for these projects. SinapisAI’s goal is much narrower than a national infrastructure plan, but it depends on the same availability of compute, data, and interoperable tools.
The Commercial Test Is Whether Agents Can Work Together Reliably
The language of autonomous collaboration can easily sound more mature than the underlying technology. Agents can already invoke tools, write code, retrieve documents, and route tasks. Yet reliable collaboration among large numbers of agents remains difficult because systems must account for conflicting goals, incomplete context, security permissions, cost controls, and evaluation. A platform that makes a capable demonstration can still struggle in a regulated enterprise environment where every action must be logged and reviewed.
That is why SinapisAI’s focus on capability characterization could become important if it can be operationalized. A system that knows a model’s strengths, limits, inputs, and expected outputs has a better chance of assigning work intelligently. But the company has not published a benchmark, a product specification, or a list of supported models. It would be premature to say that it has built a functioning marketplace or an industry-standard coordination layer.
Other Chinese startups are also pursuing the infrastructure surrounding physical AI and agents. The recent financing of DeepCtrls for physical-AI infrastructure shows that investors see value in the connective systems around models, devices, and data. SinapisAI’s ambition is more software-centric, but the investment thesis is similar: the largest opportunity may lie in the parts of the AI stack that make models usable in complex, real environments.
For now, the company’s nearly 100 million yuan angel round is a statement of investor confidence in a research-led approach to agent coordination. The funding gives SinapisAI room to develop an offering, but its success will depend on evidence that its system can cut integration time, improve task completion, or reduce the operating cost of using multiple agents. Until the startup discloses those results, its role in China’s AI market is best described as a well-funded early attempt to turn a research idea about reusable learning into working infrastructure for the agent era.