In 2025, Palantir — a company that had been around for over 20 years — suddenly became one of the biggest winners of the AI era, with a market cap that briefly touched $500 billion and stock gains that outpaced even Nvidia's. The capital markets slapped a new label on it: "enterprise AI agent benchmark."
Palantir's success has sparked a global race to copy it. In China, software companies, cloud providers, data firms, low-code platforms — basically every B2B player — is talking about Palantir and trying to follow its playbook.
But there's a lot of talk and very little real execution.
Until now. Zhongshu Ruizhi, a leading enterprise-level AI agent operating system provider in China, just closed a Series B round worth hundreds of millions of yuan. The round was co-led by Qinghua Kong Jinxin Capital, Yuanhe Zhongyuan, and Shangxianhu Fund, with existing investor CDH VGC also participating. This latest financing comes less than six months after the previous round. More importantly, from its technical approach and product lineup to business model and client base, Zhongshu Ruizhi bears a striking resemblance to Palantir. Its rapid rise is validating the "Chinese Palantir" thesis.
Why Is Palantir So Hard to Replicate?
To understand why Zhongshu Ruizhi's funding is such a big deal, you first need to understand why Palantir is so hard to copy.
On the surface, Palantir is a "four-not" company. It's not a traditional consulting firm — though it's deeply embedded in client operations. It's not a typical software company — though it ships standardized products. It's not a data company — though data processing is its foundation. And it's not a conventional AI company — though agents are its core product.
Palantir's essence can be summed up in one sentence: data must be understood before AI can truly work.
In most enterprise AI efforts, there's a chasm between data and AI. Companies have massive amounts of data and have purchased advanced AI models, but the two can't effectively connect. Data is scattered across systems with inconsistent formats, unclear semantics, and vague relationships. AI models, while powerful, can't grasp the business logic behind the data — they can only do surface-level retrieval and Q&A, not real decision-making.
Palantir solves this with its "ontology" approach. Ontology is essentially systematic modeling of a business: defining business objects (like customers, orders, equipment), clarifying relationships between them, and standardizing business rules and decision logic. Through ontology, Palantir organizes messy data into AI-comprehensible "knowledge," enabling AI to truly understand the business, make decisions, and take action. That's why Palantir can operate in high-risk settings like military command, counterterrorism, and energy scheduling — it's not about letting AI "look at data," it's about letting AI "understand the business."
This is exactly where most copycats fail. Software companies are good at building apps but lack data governance to construct a quality knowledge base. Data companies handle data well but don't understand the special needs of AI applications for reasoning and decision-making. Cloud providers offer compute but lack deep industry business knowledge to deliver end-to-end solutions. Low-code platforms lower development barriers but can't solve the problem of modeling business logic in complex scenarios.
In other words, simply imitating Palantir at the product level is useless — you need the full technical stack from data to AI application. And that's what sets Zhongshu Ruizhi apart.
As mentioned, the similarities between Zhongshu Ruizhi and Palantir are remarkable. How close, exactly?
From a technical standpoint, both focus on intelligent decision-making and data value mining in complex scenarios, both use ontology as the core to build digital twins of business logic, and both aim to make AI truly understand business, participate in decisions, and drive execution.
In terms of product portfolio, Zhongshu Ruizhi has built a full-stack product matrix covering data ingestion, knowledge organization, agent orchestration, and industry applications. Its seven core products support integrated or modular deployment, feature high standardization and low marginal costs, and can quickly adapt to complex scenarios across industries — continuing Palantir's platform product logic.
On delivery, Palantir is known for its forward-deployed engineer (FDE) model. Zhongshu Ruizhi uses an AI-FDE model, using AI plus a dynamic ontology reflection architecture to standardize the generation of business logic. This allows it to go deep into production business without heavy customization or dispatching armies of industry experts — just a few engineers for quick fine-tuning.
In terms of client base, Palantir started with defense and military clients and expanded to aviation, industrial, biopharma, and more. Zhongshu Ruizhi serves large state-owned enterprises (SOEs) in energy, aerospace, mining, defense, and biopharma — a similar focus on foundational production scenarios.
Beyond Copying
Of course, similarity is just validation that Zhongshu Ruizhi chose the right direction. The company isn't stopping there.
Zhongshu Ruizhi's real value lies in its "dynamic self-evolving ontology" technology, which represents a key leap beyond Palantir's ontology.
Part of that is a generational advantage from timing. Palantir was founded in 2003, before the Transformer architecture existed and long before large language models. So Palantir's ontology construction relies on manual definition — business experts and data analysts spend months or longer mapping business objects, defining relationship rules, and writing decision logic. Once built, that ontology is relatively static and hard to adapt quickly to business changes.
Zhongshu Ruizhi was founded in 2020, by which time Transformer architecture was mature and the dawn of the large model era was visible, making it possible to make ontology "dynamic." The company developed its own OSTARR algorithm — a multi-agent self-generation and self-optimization technology based on reinforcement learning and a reflection architecture. With this algorithm, ontology is no longer a static knowledge base but a self-evolving "digital twin of business logic."
The "dynamic" in dynamic ontology shows in three layers:
Object layer dynamic identification. Traditional ontology requires all business objects to be predefined. But in real scenarios, new objects keep emerging. Dynamic ontology can automatically identify new objects, analyze their attributes and features, and incorporate them into the knowledge system without manual remodeling.
Logic layer dynamic optimization. Business rules aren't set in stone. Market changes, policy adjustments, and technology upgrades all can alter decision logic. Dynamic ontology continuously optimizes business rules based on real-time data and historical feedback, keeping AI decision logic aligned with the latest business practices.
Action layer dynamic adjustment. AI decisions need to translate into real actions. Dynamic ontology can adjust subsequent action strategies based on execution feedback, forming a closed loop of "sense-decide-act-feedback-optimize," giving AI the ability to learn from practice.
Zhongshu Ruizhi's technical lead explained with a vivid scenario: "Suppose a satellite detects a warship disguised as a civilian merchant vessel with artillery mounted on its hull. A traditional system would identify it as a 'merchant ship' because the ontology has no category for 'camouflaged warship.' But the dynamic ontology immediately detects anomalies: the ship looks like a merchant vessel but is armed, and its route doesn't match commercial shipping lanes. The system automatically creates a new object category 'suspected camouflaged warship,' updates recognition rules, and syncs this information to all relevant systems in real time."
In such scenarios, dynamic ontology is essential. The situation changes rapidly; new equipment and methods appear constantly. You can't define all possibilities upfront — the system must be able to self-evolve.
From Oil Fields to Power Plants: Agents That Actually Ship
Zhongshu Ruizhi deeply respects and understands industry know-how — another point of philosophical alignment with Palantir.
Founder Han Han holds a Ph.D. in Electronic Engineering from Tsinghua University and was a senior visiting scholar at Columbia University. But her most prominent label isn't "academic star" — it's "technology leader who understands policy and industry best." As a core member, she participated deeply in the drafting of multiple national AI and big data policies and standards, and she's an expert on the National Information Technology Standardization Committee, ISO/IEC, and the Cybersecurity Association of China. This top-level perspective lets her accurately grasp national strategic direction and industry trends.
In 2020, Han Han founded Zhongshu Ruizhi, turning years of top-level design and industry insight into practical action to deploy AI in critical information infrastructure. She knows that getting AI into China's state-owned enterprises isn't about the algorithms — it's about embedding algorithms into extremely complex, low-tolerance business processes.
Today, Zhongshu Ruizhi has over 200 employees, 84% of them technical staff. The core team includes multiple executives with Tsinghua backgrounds and top international R&D talent. But their most distinctive label isn't "academic elite" either — it's "AI engineers who know the industry best."
At Zhongshu Ruizhi, the team's workplace isn't just the office — it's on-site at power plant stations, oil rigs, and national mission command vehicles.
In the upstream exploration scenario for oil and petrochemicals, traditional seismic exploration requires field data collection, then dozens of people spending months analyzing geological structures and locating oil and gas — a time-consuming, labor-intensive process with low efficiency. Zhongshu Ruizhi sent engineers deep into oil fields to work alongside geologists, thoroughly understanding every detail of the seismic structure interpretation workflow. The result: a seismic structure interpretation agent that can automatically process, analyze, and interpret complex data like seismic logs, well logs, and geological structure reports. It dramatically compresses work that used to take dozens of people months, significantly improving the efficiency of finding oil and gas.
In the energy and power sector, Zhongshu Ruizhi's equipment fault early warning agent combines data fabric and dynamic ontology to create a full-chain closed loop: real-time sensing from multiple data sources, intelligent fault localization, automatic generation of repair plans, and resource scheduling. This has improved fault troubleshooting efficiency by nearly 20 times, making the product a "digital twin" for customers and drastically reducing manual workload.
Amid AI anxiety, many companies rush to deploy agents that end up as "exhibit-only" — they can't integrate into real business processes. That's where Zhongshu Ruizhi is different: it doesn't just provide technical tools; it delivers value that cuts through both the technology base and the business scenario. This "penetration power" has made Zhongshu Ruizhi the first AI startup in China to deliver multiple million-yuan pure software contracts in strategic pillar industries like energy and aerospace.
This focus on real-world deployment mirrors the shift seen in tools like ComfyUI, which recently raised $30 million to turn AI prompts into actual production pipelines.
The State-Owned Enterprise Market Explodes
Just as Zhongshu Ruizhi is growing fast, China's SOE AI application market — worth hundreds of billions of yuan — is about to blow up.
In July 2025, China's State Council issued the "Opinions on Deeply Implementing the 'AI+' Action," which explicitly targets widespread deep integration of AI in six key areas by 2027, with penetration of new-generation intelligent terminals and agents exceeding 70%. By 2030, agent penetration should exceed 90%. This is the first time the national government has set clear quantitative targets for AI application penetration — a strong policy signal.
Driven by policy, SOE AI budgets are rising fast. By industry, energy, security, and telecommunications are the core tracks. Energy AI budgets have already broken through the hundred-billion-yuan mark, covering smart grid dispatching, oil and gas exploration, and new energy management. Security budgets are approaching that level, focusing on command decision-making, intelligence analysis, and equipment maintenance. Telecom is in the tens of billions, with emphasis on network optimization, customer service, and precision marketing. Other central SOE sectors contribute trillions more in market space.
Zhongshu Ruizhi's full-stack product system is highly suited to the complex needs of SOEs. Its growth trajectory proves the point.
Over the past two years, Zhongshu Ruizhi's number of major clients has doubled annually, covering dozens of top SOEs and research institutes in power, oil, aerospace, telecom, and biopharma. Repeat orders keep coming. According to Han Han, by the second half of 2026, the company expects revenue to double again. The company is already fully profitable.
Han Han said that if last year many people viewed agents as an "innovative technical experiment," now everyone agrees it's a full architectural overhaul — the inflection point has been crossed. Agent AI now has capabilities for autonomous planning, tool invocation, and end-to-end execution. It can make decisions and take actions autonomously with high precision, stability, and reliability over long periods, directing production in the physical world. In the future, agents will reshape software and processes; humans will focus on decision-making and creativity, ushering in a new era of AI-driven productivity upgrades.
With this historic opportunity of a market explosion, Zhongshu Ruizhi — with its proven full-stack product system and top-tier benchmark clients — has clearly seized the first-mover advantage. From technology breakthroughs to commercial closure, it has built a sustainable growth flywheel of "R&D → deployment → iteration → scale." This "Chinese Palantir" is writing its own growth story.