AlignBase raises hundreds of millions to build industrial AI foundation with "omni-factor large model" | Unity Ventures portfolio
Contract values have increased by an order of magnitude compared to six months ago.
Unity Ventures angel-round portfolio company AlignBase has announced the completion of three funding rounds totaling hundreds of millions of RMB.
AlignBase builds its core technology around the "Omni-Factor Large Model," focused on driving deep deployment of large models in core industrial manufacturing scenarios. Founder Zonghong Dai previously served as co-founder and VP of AI Infrastructure at 01.AI, CTO of AI at Huawei Cloud, and Director of AI Infrastructure at Alibaba's DAMO Academy.
To date, AlignBase has delivered multiple benchmark projects in key industries including metal smelting, petrochemicals, electronics manufacturing, and textiles — covering high-value scenarios such as process chain optimization, production yield improvement, supply chain coordination, inventory optimization, and production scheduling. The company is now pushing its "Omni-Factor Large Model" from single-scenario validation toward multi-industry, multi-scenario deployment.
Six months ago, AlignBase — an AI company swimming against the current into B-end customization — had just rolled out seven or eight projects. The loudest voices founder and CEO Zonghong Dai heard from the outside were skeptical: "The story isn't sexy enough," "Customization is just grunt work."
Six months later, AlignBase's response to the doubters: tens of millions of RMB in hand.
Order volume has doubled, contract value has increased by an order of magnitude compared to six months ago, and AI solutions have landed in more than 10 industries including metallurgy, chemicals, precision manufacturing, semiconductors, and textiles.
Alongside the orders came three funding rounds in six months, raising hundreds of millions of RMB. CASID, Electric Control Industrial Investment, Shanghai Semiconductor Industry Investment, Jiantou Investment, Xinheda, Chonglin Capital, and Hardcore Nut Capital have all bet on AlignBase's growth.
Zonghong Dai is a veteran in AI-to-B. He was a co-founder of 01.AI — one of the "AI Six Little Tigers" — and previously served as CTO of AI at Huawei Cloud, where he delivered hundreds of customization projects.
He has described so-called "customization" as the process of manually modeling the know-how embedded in experts and business data into a workflow.
Sorting through complex data and knowledge is precisely what makes traditional customization "dirty" and exhausting. The leap in large model reasoning capabilities gave Dai a vision for changing the customization paradigm.
What AlignBase is doing now is handing over the traditional labor-intensive, time-heavy enterprise customization service to an AI system.
The end result: transforming custom projects that traditionally required hundreds of people on-site for months into single-operator, roughly two-week deliveries — while exceeding the results of traditional big-team implementations.
Building an "Industrial World Model" for Business
After researching hundreds of enterprises, Dai found that traditional manufacturing companies don't care about office tools that boost white-collar productivity. What they care about are metrics they can actually move in production: yield rates, capacity, inventory, supply chain.
In the non-ferrous metals industry, for example, the biggest pain point is how to expand capacity while ensuring stability and safety. That's because the gains from increased capacity far outweigh the savings from pure cost-cutting.
The underlying need: enterprises need a "brain" that can iterate on business data, reference the business metrics they set, and deliver customized optimization solutions that can be directly adopted in real production.
With the improvement of large language model reasoning capabilities, Dai believes the time has come to change the traditional customization workflow: let AI directly replace customization expert teams, and based on the business metrics enterprises provide, deliver precise solutions to frontline workers.
Essentially, production and manufacturing in traditional industry can be stripped down to two questions — "what to use" and "how to do it." No matter how complex the production process, it can be broken into controllable, simple modules for large models to learn.
Traditional customization services rely on expert experience modeling, consuming heavy human and time resources, and struggle to respond to enterprises' needs for global optimization and flexible change. In Dai's view, large models' learning and reasoning capabilities can precisely solve the problems of difficult modeling and low efficiency.
Another benefit of replacing experts with models: the ability to mine every potential optimization point in production, rather than just single-point optimization for one link.
Therefore, AlignBase developed its own industrial AI operating system, called the "Omni-Factor Large Model," serving as the central brain guiding enterprise production operations. Its logic runs in three steps:
Learn: Using enterprises' raw business data, comprehensively learn the business model and establish a digital twin that reflects real production processes.
This model is the underlying architecture of the entire system. It continuously updates with new data injections, enables precise information tracking, and filters out noise from unreliable or missing data. This is because the model can mine intrinsic correlations between data and focus attention on data that has greater impact on key production metrics.
Optimize: As enterprise data continuously improves and the model's own reinforcement learning capabilities strengthen, the system continuously simulates and searches for optimal solutions in production workflows.
Deliver: Directly facing frontline workers, delivering an app that can interact with the AI system. The app's interface and operations are extremely simple — workers only need to input the on-site environment to receive the current optimal production plan. In metallurgy scenarios, for example, the system tells workers how much material to stack, when to stack it, and how to stack it.
△ Omni-Factor Large Model, image source: AlignBase
Dai calls this system an "industrial world model" about data and business.
Dai believes business scenarios are also part of the world — human business decisions are essentially predictions of future business situations based on business data.
So they have built a world model for industrial scenarios, projecting business scenarios into the digital world, aiming to learn correlations between data to find which factors influence each other and how, thereby predicting and guiding actual production optimization.
To put it another way: on a real production line, this system can learn and analyze an enterprise's existing business data, map out and replicate the production process, and generate a virtual twin "digital factory" model.
It can continuously run self-simulations, identifying better practical operation plans according to the production metrics the line requires, which frontline employees can directly implement.
△ AlignBase product architecture, image source: AlignBase
Dai summarizes their approach as "improving quality and efficiency" rather than "reducing headcount to improve efficiency." The advantage: they don't create digital workers, they don't use AI to replace existing human labor.
They found that when vendors propose replacing real people with AI projects, these traditional enterprises are actually less receptive — partly because human labor is much cheaper than AI projects, and partly because they expect short-term, visible capacity improvements rather than long-term headcount reduction.
Under enterprises' existing production operation models, AlignBase uses system-provided solution design to address the pain points enterprises truly care about — "quality" metrics like output value and yield rate — achieving efficiency gains across the entire production line.
In terms of delivery results, on a single process segment, their system can help improve a key metric by 2–3x, with annual cost savings reaching tens of millions of RMB.
No Pie-in-the-Sky Promises, Directly Delivering Optimization Metrics
For its first landing spot, AlignBase didn't choose the more digitally mature internet industry. Instead, it first deployed its AI system in so-called traditional industries: metallurgy, chemicals, precision manufacturing, semiconductors, textiles.
In the public eye, traditional industry is a tough track — not a sexy story, and enterprises' data governance, their management of their own data assets, isn't strong enough.
Dai gives an answer that runs counter to stereotypes: "In my view, industrial enterprises are actually easier to work with." His reasoning is simple: traditional industries are quite large, making it easy to achieve scale effects.
The reason for not choosing the internet industry is that it's natively digital — internet industry solutions require more disruptive innovation, which places higher demands on customization vendors.
In actual production, industrial enterprises have all kinds of raw business data: logs, operation logs, ERP data, PRD data. These data come in different formats, are noisy, and even have missing parts — seemingly tough nuts to crack.
Dai doesn't see this as a problem. In fact, AlignBase doesn't particularly need enterprises to perform complex data governance themselves.
"Governed data is like chewed food — it loses much of its original information," Dai explains. The direct business data in enterprise systems retains more complete information from production and manufacturing processes, which is more helpful for them.
In the process of building clients' "industrial world models," all data AlignBase uses comes from the clients themselves, with no reliance on know-how experts. This also makes the system easy to migrate and deploy across industries.
In the higher-willingness-to-pay ToB domain, AlignBase inevitably faces off against big tech and established industry solution providers. The team's customer acquisition strategy: making business optimization metrics a mandatory deliverable, written directly into contracts.
Because delivery outcomes are uncontrollable, most competitors dare not write specific, guaranteed business metrics into contracts. This leaves clients' business pain points unresolved. In contrast, based on the "Omni-Factor Large Model," AlignBase can precisely deliver guaranteed business metrics and corresponding optimization plans according to client pain points.
Meanwhile, AlignBase employs a clever pricing model: pricing by expected outcomes, not actual delivery outcomes.
"If we priced by actual delivery results, clients would desperately lowball the metrics before delivery," Dai explains. "We want to build a win-win relationship with clients, not fight each other for money." To get clients willing to pay for expected outcomes, AlignBase promises "minimum delivery metrics" for business improvement in contracts.
Despite solid commercial results, Dai candidly told us that AlignBase's AI solutions are not yet generalized enough.
For example, current clients are mainly leading enterprises with higher data governance standards, so AI solutions have not yet fully generalized to production scenarios at small and medium enterprises.
To improve system planning capabilities and generalizability, AlignBase plans to first切入 5–10 industries, solidifying single-industry deployments before eventually generalizing to more industries.
Dai noted: "Next, we need to achieve end-to-end delivery across even larger industries, produce at least two standardized products, and achieve actual delivery."