Former 01.AI Co-founder Dai Zonghong's Startup "Basis Origin" Closes Angel Round of Over 100 Million Yuan | New Unity Ventures Portfolio Company
A 20-person team completes AI customization projects that previously required hundreds of people
Recently, BasePoint Origin completed an angel round exceeding RMB 100 million. Investors include Unity Ventures, Innovation Works, JianGuo Capital, Puhua Capital, Yinqiao Fund, Zhengyang Hengzhuo, CASSTAR, and Dreame Ventures.
BasePoint Origin was founded by Zonghong Dai, former co-founder of 01.AI, and aims to build an "AI operating system" for enterprises. This system is a technical platform integrating models and applications. Its foundation comprises multiple foundation models and industry-specific vertical models, while the upper layer uses enterprises' own business data for modeling, ultimately combining into a virtual-physical fused digital twin environment to complete industrial AI transformation and enhance business value.
Dai believes domestic and overseas enterprises differ in payment preferences. Overseas enterprises are accustomed to paying for tools, while domestic enterprises prefer paying for end-to-end customized solutions—and their budgets are substantial.
Currently, BasePoint Origin's AI operating system is already running at enterprises across different industries including steel smelting, environmental governance, and new energy management, helping them achieve cost reduction, efficiency improvement, and supply chain optimization.
Source: Intelligent Emergence
Author: Xinyu Zhou
Domestic market, ToB, customization—these don't seem to be the buzzwords in AI entrepreneurship right now. But this is precisely what Zonghong Dai, former co-founder of 01.AI, is doing now.
On March 24, 2025, after leaving 01.AI, he established his new company "BasePoint Origin." Entering B-end customization now, Dai wants to tell a fundamentally different story from traditional B-end customization.
This story is currently reflected in BasePoint Origin's operations: with seven or eight customization projects running in parallel, the engineering team maintains a scale of roughly 20 people, and has never missed a delivery deadline.
The key lies in using AI to automate the "dirty work" in the customization process.
Traditional customization requires AI companies to assemble expert teams of dozens of people, conducting manual interviews and data collection upfront, then manually analyzing and modeling the data, and finally digitally replicating business processes—this process is essentially building a digital copy of real business manually.
Based on this "copy" of business workflows, enterprises can then simulate solutions according to their defined business objectives, and finally find the optimal path.
What BasePoint Origin aims to do is hand over the entire process of expert interviews, data collection, analysis, modeling, and business workflow characterization to AI.
To this end, BasePoint Origin has built an AI operating system that can automatically characterize enterprise business workflows:
-
The system's foundation consists of multiple foundation models and industry vertical models. Based on various raw business data from enterprises, the underlying models can automatically sort out and understand the production factor nodes of the business.
-
The built-in production factor toolchain can model each production factor individually, forming business nodes.
-
The RL toolchain within the system, based on digital twins of real business workflows, builds customized enterprise large models and other AI software through reinforcement learning.
After adopting the AI operating system, enterprise data collection and governance, plus building enterprise business workflows, takes only one day, "and after our clients review it, there are no errors."
Today, this AI operating system is already running at enterprises from different industries including steel smelting, environmental governance, and new energy management, helping them achieve cost reduction, efficiency improvement, supply chain optimization, and other goals.
For example, when an enterprise's business objective is to optimize supply chain costs by 15%, they only need to connect their supply chain data API to the AI operating system, and they can obtain their business model in a short time. After setting the 15% optimization target, the business model will provide a path to achieve the goal based on business knowledge.
Recently, BasePoint Origin completed an angel round exceeding RMB 100 million. Investors include Innovation Works, JianGuo Capital, Unity Ventures, Puhua Capital, Yinqiao Fund, Zhengyang Hengzhuo, CASSTAR, and Dreame Ventures (in alphabetical order).
Dai told Intelligent Emergence that during the fundraising process, some investors believed Chinese B-end clients have low willingness to pay. Indeed, payment cycles and client payment willingness are perennial topics.
As a veteran of AI ToB, Dai has a different view. Previously, he served as AI Infra Director at DAMO Academy, and later became AI CTO at Huawei Cloud.
During his time at Huawei Cloud, he engaged with hundreds of AI customization projects. This made him realize: "It's not that domestic willingness to pay is low, but rather that payment preferences differ from overseas. Overseas enterprises are accustomed to paying for tools, while domestic enterprises are accustomed to paying for results."
This means the domestic B-end market still has massive demand for AI customization.
After starting his business, Dai spends nearly 40% of his time on client visits. He found that these clients are unwilling to use off-the-shelf products on the market, because they cannot directly integrate with existing business, "they still prefer to pay for end-to-end customized solutions, and their budgets are very high."
In conversations with clients, he realized, "most AI companies in the past simply deployed a large model for enterprises, but what clients want isn't the large model—it's results."
Therefore, at the customer acquisition stage, BasePoint Origin doesn't rush to have enterprises deploy its AI operating system. Instead, based on the client's business objectives, it directly delivers a solution run through the AI operating system.
Dai believes that overall, competition is a good thing for the B-end, "once there are more successful cases, client confidence will strengthen, and customer acquisition will become easier for us."
The following is a conversation between Intelligent Emergence and Zonghong Dai, edited and organized.
Bridging the Gap Between Industry and AI with an AI Operating System
Intelligent Emergence: How would you describe BasePoint Origin's business?
Zonghong Dai: I hope to provide what I call an "AI operating system," a technical platform. Based on this platform, we can bridge the gap between industry and AI, allowing the core business chains of industries to quickly leverage existing AI tools and methods to complete industrial AI transformation, ultimately enhancing business value.
Intelligent Emergence: Can you give an example?
Zonghong Dai: For example, from a client's various systems' raw data, our models and platform self-learn this data, and this learning process is fully automated.
We use a few machines, learning for roughly a day or so, and we can automatically write out the enterprise's entire business process.
In traditional customization, if people want to characterize enterprise business, it generally requires interviews with personnel at various levels, technical interviews, and data sorting and governance. Finally producing results takes six months to a year.
But we can now clearly sort out the entire business in one day. And after detailed client review, there are no errors.
Intelligent Emergence: How is this different from traditional customization?
Zonghong Dai: Previously, almost all enterprises serving industries relied primarily on customization and hand-tuned models, unable to avoid data governance issues or business workflow learning problems.
But our current technology can leverage enterprises' complex data to self-learn and understand these things, then reproduce enterprise workflows. This is the unsupervised data governance technical system we've built.
Intelligent Emergence: What are the benefits of this technical system for customization?
Zonghong Dai: First, it can generalize. Second, it can penetrate into core business chains, not just serve as an office assistant. Third, related to the first, it can achieve scale, rather than just helping one or two leading enterprises complete transformation.
Intelligent Emergence: You became a co-founder of 01.AI during the 2023 large model wave, and were among the first to dive into large model entrepreneurship. Did you see the opportunity for AI customization at that time?
Zonghong Dai: From ChatGPT itself, I couldn't see the opportunity to empower thousands of industries—it stood out more for its linguistic and logical capabilities. Only when models have deep reasoning and thinking capabilities can they understand enterprise business workflows and assist in decision-making.
It wasn't until o1's release that I felt large model deployment across thousands of industries became a question of optimization degree, not feasibility.
In fact, when I was at 01.AI, we also tried doing work similar to ReAct (Reason+Act, an agent decision-making mechanism), and achieved certain results.
Though not as impressive as o1. But we made considerable progress, which we achieved independently without guidance from o1 or o3. Because we had been exploring this layer of foundational technology.
Intelligent Emergence: Why did you choose to leave Huawei in 2023 and join 01.AI in the large model field?
Zonghong Dai: At Huawei, I was able to participate in and even lead some AI empowerment work across thousands of industries, gaining many practical opportunities and exposure to numerous enterprises and scenarios. So this experience was indispensable for me, including some prototypes of my current entrepreneurship that took shape at Huawei.
But at Huawei, it was difficult for me to delve into large model technical details. At that time, 01.AI already had preliminary technical foundations and certain funding foundations, and finally I could clearly see that I could accomplish many things there. So after talking with Kai-Fu and Xuemei, I joined.
At 01.AI, I systematically observed and learned about a series of large model technologies. Because previously I was assisting with AI, but the experience of deeply immersing myself in driving every metric at 01.AI was quite different.
Intelligent Emergence: Which technological developments became the catalyst for your entrepreneurship?
Zonghong Dai: I'm not a large model technical expert myself. The reason I started working on large models is because I believe they can help me achieve the three points above.
I believe deep Reasoning capability is a critical inflection point, which was when OpenAI's reasoning models o1 and o3 were released in 2024. The reason I launched the "BasePoint Origin" project at this time is also because large models had reached this level.
I feel I don't need to do much work on large model training myself—I can directly stand on others' shoulders.
20-Person Team Doing the Work of Hundreds at Traditional AI Companies
Intelligent Emergence: What is the structure of the current "AI operating system"?
Zonghong Dai: It's essentially a platform integrating models and applications. The bottom layer of our system is a set of large models, including foundation models and industry vertical models, serving as tools for understanding business data and modeling.
The upper layer of the system uses enterprises' own business data for modeling, ultimately combining into a virtual-physical fused digital twin environment. These built models have no hallucinations because they're based on the enterprise's business data, so they understand the enterprise's business chain.
Intelligent Emergence: What is the method to achieve "using AI to characterize complete business workflows"?
Zonghong Dai: An enterprise's data is very complex, with different sources and storage formats. What we first need to do is automatically and deeply mine this data, including mining its latent representations, and perform automatic data analysis.
Next, we need to mine what we call the "complete set of production factors" based on this existing data, then build these into business operations.
Then we need to perform automated modeling on each node. Automated modeling relies on data. For some nodes without natural data, we need to perform automated data completion.
In short, it's presenting an enterprise's real business process through modeling.
The whole process sounds very simple, but achieving automation is very complex. Because some enterprises have tens of thousands or even millions of nodes, we need to automatically learn the relationships between these nodes, including developmental and logical relationships, which is quite challenging.
Intelligent Emergence: How do you determine that the completed data conforms to real business logic?
Zonghong Dai: We don't need it to be precise, we just need it to be correct. Because later we'll actively converge model accuracy through reinforcement learning.
Essentially in the AI 1.0 era, people did more work improving model accuracy through labeled data. This required massive human effort, and the uncertainty and instability were very high.
We no longer use manually labeled data. We need to form natural reinforcement learning environments through clever design, continuously producing data through actual production environments to strengthen our models.
Essentially, this process can achieve equivalent effects to manual data labeling.
Intelligent Emergence: Why use AI to model enterprise business workflows? What value does this bring to enterprise clients?
Zonghong Dai: Enterprises can use this characterized business workflow to conduct business simulation and decision-making.
You can think of our replicated business workflow as a digital transformation environment, a mirror image of the real business workflow. Doing various simulations in this digital environment is relatively easy.
Simulations based on the digital environment also bring many possibilities for enterprise business implementation. For example, besides helping enterprises simulate effects of different decisions, it can also do cost reduction, efficiency improvement, production capacity enhancement, production capacity balancing, supply chain optimization, and so on.
Intelligent Emergence: Does this process require massive human investment like traditional customization?
Zonghong Dai: The degree of human intervention in the middle is almost zero. It's like when a client comes, as long as we obtain permission for the corresponding data interface, the entire model learning and understanding process is fully automated.
Intelligent Emergence: Is the final effect of this system the same as traditional manual customization?
Zonghong Dai: What this system replaces isn't actually some internal enterprise functions, but rather AI company experts helping an enterprise with intelligentization in certain scenarios.
According to traditional customization processes, we would need to dispatch large numbers of experts to help enterprises with corresponding analysis and modeling work. We are now essentially using large models to replace these AI company experts, helping enterprises with complete-factor modeling.
If relying on people, 10 people can only build 10 models, 20 people can only build 20. If an enterprise has 10,000 production factors that all need modeling, you would need 10,000 models, corresponding to 10,000 people.
But with this system, we now don't rely on human power—we rely purely on computing power.
Intelligent Emergence: Many companies doing ToB customization eventually become bloated in personnel scale. How many people does BasePoint Origin have?
Zonghong Dai: We're currently pushing seven or eight projects simultaneously, but the responsible team totals only 20 people—previously this might have required over a hundred. So far our delivery times have never failed to meet client expectations.
This is a completely different organizational form from traditional customization delivery. The reason is simple—we've solved most problems through technology.
Intelligent Emergence: What stage has this AI operating system reached?
Zonghong Dai: In early August, we just completed delivery of the first version of core functions, so it also went through three to four months of construction time.
Intelligent Emergence: How is this AI operating system actually performing in enterprise deployments?
Zonghong Dai: Our results are already very helpful for enterprises. Enterprises only need to let us read historically accumulated data, and we can quickly characterize their complex business chains.
Our enterprise cooperation cases already include using replicated digital business workflows for cost reduction and efficiency improvement, supply chain optimization, and some enterprises have done energy consumption management optimization.
Intelligent Emergence: How should enterprise clients operate such a platform to achieve business objectives?
Zonghong Dai: For this part of the product, we haven't completed end-to-end automation yet. We currently manually string together different capability nodes, because our development time hasn't been that long. Now when we obtain client data, we manually string together each node that AI automatically models.
Clients give us their defined business objectives, such as "achieve fastest production speed" or "achieve maximum production capacity," and our system automatically helps users optimize and find the path to achieve maximum objectives.
Domestic Enterprises Prefer Paying for Results
Intelligent Emergence: Is the company currently targeting the domestic market or overseas market?
Zonghong Dai: Primarily the domestic market. Undoubtedly, China has the advantage of a complete manufacturing industry chain. For solutions like ours, the longer the industry chain we can connect in the future, the greater the optimization space. So China actually has natural scenario advantages.
Overseas there may be some enterprises with stronger single-point capabilities, but end-to-end complete industry chains are relatively weaker compared to China.
Intelligent Emergence: Now that you're doing ToB and domestic, is there pressure in fundraising?
Zonghong Dai: We met investors with aligned philosophies.
When meeting investors, I have very clear feelings. If investors have previously visited factories for projects, or even operated some things hands-on, they very easily understand when we talk about business and describe our technical path—basically succeeding in one conversation.
Before we realized this technology, some investors were skeptical about whether AI automation of process characterization could be achieved. I quite understand this point.
We were fairly lucky, quickly finding investors who could understand our technical philosophy and some philosophical thinking, and quickly finalized the funding. The angel round went quite well—our funding amount has already exceeded 100 million.
Intelligent Emergence: Are you worried about low payment willingness for ToB software among domestic enterprises?
Zonghong Dai: I don't think domestic payment willingness is low—rather, payment preferences differ from overseas. Overseas enterprises are accustomed to paying for tools, which is something excellent overseas, ultimately able to sustain these tool-making enterprises.
Domestic enterprises are more willing to pay for results. Most enterprises are willing to pay for end-to-end business value, and their cooperation willingness is very strong.
For the projects we're currently doing, enterprises' payment willingness is very clear, and their budgets are also ample. If the product can help them improve efficiency throughout the entire process, and help clients save money or increase revenue, they will definitely be willing to allocate budget to cooperate with us.
The clients we're currently serving clearly demonstrate this characteristic—they include central state-owned enterprises and private enterprises.
Intelligent Emergence: How many clients does the company currently have?
Zonghong Dai: We already have nearly 10 projects running simultaneously, and they're from thousands of industries. Their payment willingness is also very strong, with very high amounts of proactive advance payments.
Intelligent Emergence: How should enterprise clients measure the effects produced by the AI OS? How should pricing and payment work?
Zonghong Dai: In traditional customization, to put it more directly, software is essentially a data dashboard. A data dashboard requires people to understand data and make decisions to generate value. At this point, the value of a data dashboard is very difficult to measure.
But what we're doing now is fully automated AI, so the value produced can be simulated.
Second, previously most enterprises were doing single-point optimization. How much effect single-point optimization can produce when placed in the global context—you have to implement it and execute it to measure many things. So pricing was difficult before.
What we're doing now is a complete-factor model, containing every node end-to-end, so the value produced by single points in the global context, we can all measure.
Intelligent Emergence: Can this business model break even?
Zonghong Dai: If we don't invest in future R&D, we have hope of breaking even this year.
Intelligent Emergence: Would you consider first doing some business with quick commercial returns to sustain the company, then using that money to sustain your ideals?
Zonghong Dai: The core of commercialization should be helping clients create their value, then taking a share from their value to sustain ourselves. Generally, we hope to help clients achieve 10x returns before obtaining our own returns—this is a virtuous cycle. Otherwise it's just a random project.
Intelligent Emergence: If random projects can sustain the team, isn't that also a kind of value?
Zonghong Dai: I think this may be harm to the industry. For example, if I promise a client to create 30 million in value, but in the end it's not even 10 million. This was originally a client willing to try new technology—after their trust is damaged, next time they may not even be willing to pay 10 million.
Market confidence is very important, especially in B-end, because B-end memory is very deep. This year's performance will directly affect their budgets for AI technology in the coming years.
Intelligent Emergence: Have you encountered clients who have been hurt before? How do you help clients rebuild confidence in the market?
Zonghong Dai: There have been some. Clearly their sunk cost tolerance is much lower than enterprises that haven't been hurt. Other enterprises can accept seeing business value in about half a year—these enterprises, if they don't see it in about three months, are less willing to invest. So we need to calculate data very meticulously, with business value judgments for enterprises that can withstand scrutiny.
Intelligent Emergence: No one has done such an AI operating system replicating business before—how do you make clients understand?
Zonghong Dai: Many traditional AI companies find it difficult to break through with clients because they're still telling stories about intelligent assistance, or building a large model for enterprises. But enterprises don't care whether they have a large model—they care what business value they can achieve after having one.
If Chinese AI companies can take one more step forward, and very clearly help enterprises sort out the presentation of business value after adding AI, breaking through enterprise clients is relatively easy.
We were fairly lucky, catching the wave of China's AI+ transformation, plus our solution and technical path can present business value, so breaking through with clients wasn't that difficult.
Intelligent Emergence: For such an AI OS, are there industries or enterprises where deployment effects are particularly good?
Zonghong Dai: Actually this AI OS is applicable across all industries—it's essentially a general operating platform.
We already have several enterprise clients, with decent delivery. They include metal smelting and steel smelting, environmental governance, new energy management, and electronics manufacturing and assembly—quite a broad span.
Competition Will Increase Successful Deployment Cases
Intelligent Emergence: What are the barriers to your team's business?
Zonghong Dai: First, we have deep understanding of industries. Second, we also have deep understanding of model development itself. Third, the technical and product system we've designed is forward-looking, and our frontier breakthrough capability is quite good, able to gradually achieve according to our plans.
This means we have good cognition, appropriate design, and strong practical capability.
Intelligent Emergence: Many enterprises mention "industry know-how" as a barrier. What is true "industry know-how"?
Zonghong Dai: During my time at Huawei, I accumulated experience from hundreds of AI+ industry empowerment projects.
At that time we were still using traditional AI technology, but in that process we gained deep cognition about what enterprises truly need, industry work characteristics, existing system data in industries, value chains, and so on. And these cognitions were simulated through projects one by one over several years.
So to do what I'm doing now, there are several key points: first, we have deep understanding of industry work patterns and core managers' thinking;
Second, we also need deep understanding of AI technology. We need to know where today's AI capability boundaries are, and also know approximately what state AI will develop to in half a year, one year, two years. Otherwise, solutions designed and launched based on today's AI capabilities may need to be completely overhauled in the future.
We also need deep understanding of costs, computing power, business layout, and sensitivity to user data. So this is a very complex system.
Intelligent Emergence: How do you predict AI capability boundaries one or two years out?
Zonghong Dai: Actually it doesn't need to be that precise—as long as we have clear judgment on general direction and approximate capability boundary range.
First, I think next year large model capabilities themselves will be difficult to break through the data barrier. Because today natural data has been consumed too much, and now more data is iterated further using the approach of lifting oneself by one's own bootstraps.
Due to this point, large models will find it difficult to form fundamental breakthroughs within 1-2 years. So models' general knowledge capabilities will have linear increases, but relatively difficult to have order-of-magnitude increases.
Second, models' reflection on specific scenarios in terms of thinking chain connections, self-correction and self-iteration capabilities, and self-evolution capabilities through reinforcement learning methods—I think there will be tremendous evolution in the next two years.
Intelligent Emergence: How do these two judgments specifically translate to BasePoint Origin's current work, and what role do they play in product and business progress?
Zonghong Dai: Essentially it makes me feel that we cannot rely on researching AI capabilities themselves to solve industry problems, because AI capability iteration has slowed. Problems must be solved through business layout, not industry data.
If we want to use business data to solve industry problems, then its presentation form itself is not a simple vertical large model, but a complex DAG-based model combination and logic relationship graph. This is the foundation of our complete-factor model construction.
Intelligent Emergence: Any regrets from two years of large model entrepreneurship?
Zonghong Dai: When I was at 01.AI, the team was excellent. In June-July 2024, several of us experts together explored deep thinking capabilities, and at that time the model could already make decent inferences.
But at that time one inference took over ten minutes. We thought if a simple inference took that long, complex inference would take even longer. Actually we could have gone further in thinking chain and reinforcement learning investment and research.
Intelligent Emergence: What experience does this bring to this entrepreneurship?
Zonghong Dai: I think the simplest thing is combining with real scenarios, making joint judgments with future scenarios. This is the most direct judgment of technical value.
Second is making more judgments from technical principles, rather than based on one or two metrics.
Intelligent Emergence: What important technical progress has there been for you this year?
Zonghong Dai: Currently there haven't been key results strongly combined with our business. I'm hoping to see models that don't rely on short-term memory, but can form self-feedback learning based on scenarios. This is hopeful to achieve, and we're also doing work in this area ourselves.
Intelligent Emergence: Are you worried about competition?
Zonghong Dai: Not worried. First, these cognitions, understandings, and technologies—my team and I have accumulated them for nearly ten years. In single capabilities, there are many people no worse than us, but as a composite team, we still have certain scarcity.
Second, the market we face is large enough. In China's industrial sector, there are 512,000 above-scale enterprises, and these enterprises all have the capability and willingness to do intelligent transformation.
Actually domestically, we are the first enterprise to do fully automated business modeling. I also hope to see more enterprises like us emerge—this will help more traditional Chinese enterprises quickly upgrade to new quality productive forces.
Overall, competition is a good thing. Once there are more successful cases, client confidence will strengthen, and customer acquisition will become easier for us. If this market space were very small, one or two enterprises dying would mean others couldn't survive either.
