Artificial intelligence is no longer just an application tool; it has begun to help humanity tackle critical "bottleneck" problems in fundamental science.

"If the company is a ship exploring the deep sea, I am the one who absolutely cannot jump ship," said Jia Haojun, founder of Depth Principle, whose headquarters are located in Hangzhou. Jia has named his office "Columbus." In his view, starting a business in the brand-new field of AI for Science is akin to "Columbus exploring a new continent."

In the more than a year since its establishment, Depth Principle has been rapidly advancing in research and development, financing, and commercialization. Jia Haojun maintains five to ten minutes of deep thinking every morning to assess the company's current risks and identify the next objectives. This habit started in 2023 when Jia began his entrepreneurial journey while pursuing his PhD at the Massachusetts Institute of Technology (MIT).

AI for Science (abbreviated as AI4S in the industry) refers to using AI to make new scientific discoveries. In 2023, the Baker Institute team in the US, in collaboration with Google's DeepMind, developed the deep learning model "RFdiffusion," which predicted approximately 200 million protein structures and could design and generate proteins with a single click. In 2024, the Nobel Prize in Chemistry was awarded to the Baker Institute and DeepMind teams.

In the same year, Jia Haojun officially founded Depth Principle. The team applies AI to materials research based on the integration of generative AI and first principles. To date, Depth Principle has independently developed six major algorithm modules, integrated into a self-developed platform called "ReactiveAI." Recently, the platform was upgraded to the Material Discovery Agent (Agent Mira), which can autonomously mobilize data and resources for chemical materials R&D based on client requirements.

The year 2025 marked a critical inflection point for AI4S. In August, China released its "AI Plus" initiative, identifying AI4S as a key direction for upgrading the paradigm of scientific discovery. On November 25, former President Trump signed the "Genesis Plan" executive order, elevating the use of AI to transform scientific research methods to a national level in the United States. Around the same time, hundreds of AI4S startup projects emerged in Silicon Valley.

Subsequently, US National Laboratories, OpenAI, and DeepMind continued to increase their investments in AI4S. On January 12, 2026, Nvidia and Eli Lilly announced plans to invest $1 billion over five years to establish a joint research laboratory in San Francisco for AI drug discovery. On the same day, Anthropic announced the launch of healthcare and life sciences services to help Claude users share health records. General-purpose scientific research engines like Kosmos and Biomni were also released successively.

Domestic tech giants responded swiftly. Tencent established a life sciences laboratory in September 2025; Alibaba promoted the development of the LucaOne large model, the industry's first biological large model integrating DNA, RNA, and proteins; ByteDance specifically set up an AI for Science team integrated into its Seed department and co-established an "AI + High-Throughput Joint Laboratory" with BYD for lithium batteries.

"Using AI merely for chatting or generating videos is underutilizing its potential," Jia Haojun stated. He believes AI's most valuable contribution lies in empowering exploration into unknown human domains, asserting that "all scientific progress is essentially driven by new discoveries."

In November 2025, Depth Principle completed a Series A funding round exceeding 100 million RMB. The round was co-led by the Alibaba Entrepreneurs Fund Greater Bay Area Fund, managed by Gobi Partners, and Ant Group, with participation from existing shareholders including Lenovo Capital, Taihill Venture, and BV Baidu Venture.

A managing director from Lenovo Capital commented: "The world has entered a stage of co-innovation combining 'scientific discovery + super technology engineering.' AI4S can precisely solve critical bottleneck problems in fields like high-end manufacturing and biopharmaceuticals—such as delays in new materials R&D and excessively high costs for innovative drug development—shortening research cycles from years to months or even weeks, directly enhancing national competitiveness in basic scientific research."

Despite being on the eve of a potential value explosion, Jia Haojun still contemplates the company's direction from a pessimistic angle each morning, stemming from the many uncertainties surrounding AI's empowerment of scientific research.

On one hand, the field of scientific discovery, particularly in industry, has been relatively closed and conservative historically, with data access limiting the development of vertical-specific models. On the other hand, the standardization and digitization processes in chemical materials R&D are still underway, leading to deficiencies in the quantity and quality of relevant historical data. In other words, the data foundation for AI-driven scientific discovery is very weak, and collaboration mechanisms are underdeveloped.

However, from Jia Haojun's perspective, beyond competing with large corporations, he places greater importance on how Depth Principle can rapidly achieve the industrial application of AI4S, creating tangible value from the technology they possess.

The technological timing is now opportune. Since childhood, Jia Haojun has been fascinated by two things: science and computers, which became the origin of his entry into the AI4S industry.

In 2015, Jia chose to major in physics for his undergraduate studies. By his sophomore year, he was already using CPUs for first-principles calculations based on the Schrödinger equation. However, at that time, computational limitations made the process extremely time-consuming and labor-intensive, with almost zero commercial potential. "Submitting a calculation task would require a supercomputer to run for several days."

This changed in 2018. GPU computing power advanced by leaps and bounds—computational efficiency increased tens of times, reducing molecular calculations that previously took days to just minutes. Simultaneously, neural networks began to be widely applied, with AI starting to be used to "predict" molecular behavior patterns instead of relying solely on brute-force calculation.

This made Jia Haojun aware of AI's vast potential in chemical materials R&D. In 2019, upon starting his PhD at MIT, Jia proactively applied to change advisors, seeking professors conducting research in this direction. He eventually studied under MIT Chemical Engineering Professor Heather Kulik, a leading figure in AI chemical design.

Heather Kulik and Nobel Chemistry laureate John Jumper, a core member of DeepMind, were among the earliest global scholars using artificial intelligence for scientific discovery. John Jumper's research focused on AI predicting protein structures, while Heather Kulik applied AI algorithms to the field of chemical materials discovery.

At that time, Jia's senior labmate, Duan Chenru, had already been researching AI for Materials for a year. Duan focused on underlying AI algorithms and computational methods, while Jia specialized in material application transformation and reaction systems. Their research分工 resembled a combination of "principle" and "application." This division of labor continued into the founding of Depth Principle: Duan Chenru is responsible for technical architecture and algorithm R&D, serving as CTO, while Jia Haojun handles strategy, clients, and team building, serving as CEO.

In Professor Kulik's evaluation, Duan Chenru possesses "outstanding academic leadership" in the AI4S field, while Jia Haojun was "the bravest student" when facing complex research challenges. According to an angel investor and partner at Linear Capital, Zeng Yingzhe, "The two of them starting a business together is a perfect partnership."

Although Jia Haojun successfully joined an AI4S research group, at that time, AI was primarily seen as a "tool," and its applications quickly hit a ceiling, with the industrialization of AI4S remaining unclear.

"The mainstream approach then was using AI to accelerate traditional processes—calculating faster and fitting more accurately," Jia explained. "Many people get hit by apples, but only Newton proposed the law of universal gravitation. Most others spend their time verifying the new discoveries proposed by that tiny minority. AI became a research tool, but the scientific research paradigm itself hadn't changed: 99% of scientists were still engaged in verification work, with the hypothesis generation phase still relying on human intuition."

A turning point occurred at the end of 2022. The emergence of ChatGPT made Jia Haojun realize that "generative AI" represented a qualitative leap compared to previous AI. Consequently, Jia and Duan began very close academic collaboration, jointly researching how to apply generative AI technology to chemical materials R&D. Throughout their PhD careers, Jia and Duan published over 60 papers collectively in top-tier journals and platforms like Nature sub-journals, pioneering several new models for AI for Materials.

At that time, the industry predominantly used large language models (LLMs) in areas like molecular synthesis and drug discovery. However, the core challenge in chemical materials R&D is not a "lack of knowledge" but a "lack of verifiable candidate structures." An LLM might write an essay about catalysts but cannot directly generate a computable molecular coordinate file.

Therefore, in the chemical materials field, differentiated models are needed to supplement LLMs' understanding of structures, leading them to consider diffusion models. The latter output structured data that can directly interface with simulation and experiment.

Furthermore, chemical reactions involve many-body systems requiring consideration of issues like symmetry, which traditional SE(3)-equivariant diffusion models struggled to address. Jia and Duan decided to develop their own graph neural network to ensure symmetry in chemical reactions, simultaneously combining this network with a diffusion model architecture to create a system capable of generating complete chemical reactions.

In 2022, their work achieved a major breakthrough—they became the first team globally to verify that diffusion models could directly generate chemical molecules and reactions. In 2023, research published as the cover paper in *Nature Computational Science* demonstrated that new chemical reactions could be generated in seconds, compared to weeks required for manual derivation using traditional methods.

Diffusion models were originally a technology from the image generation field. Applying them to molecular generation signified AI's shift from "predicting the known" to "exploring the unknown." Reportedly, Depth Principle is currently advancing two generative AI paths simultaneously: diffusion generative models and large language models.

But challenges followed: How to ensure the physical feasibility of generated material structures? How to guarantee that materials can actually be synthesized?

Depth Principle's solution is to construct a "hierarchical generation" architecture: First, using a diffusion model at the base layer to generate coarse-grained structures, akin to sketching a molecular "draft." Second, refining details based on precise calculations using first principles from quantum chemistry. Third, combining high-throughput experimental validation at the top layer to test stability, i.e., using automated experiments to check if the "draft" is viable.

This "AI model prediction - computational support - experimental verification"流程 enhances computational efficiency by hundreds of times. "Our latest model can generate and screen thousands of candidate materials within minutes," Jia Haojun stated, "whereas traditional high-throughput calculations require months."

This closed loop is currently named the "ECML system," which Depth Principle also refers to as the "fifth paradigm for AI materials R&D."

Algorithm, data, and computing power are the three key elements of AI, with algorithms being the main battleground for AI companies to build their moats. Based on this, Depth Principle has self-developed six major algorithm modules: ReactGen (molecule generation), ReactBO (broad screening), Reactify (precision calculation), ReactControl (resource scheduling), ReactNet (synthesis navigation), and ReactHTE (high-throughput experiment). These six modules cover the entire process from R&D and synthesis to verification of a new material.

In short, Depth Principle empowers the entire chain of material R&D, synthesis, and verification with large model technology, forming the closed loop that constitutes the ReactiveAI platform.

While most scholars take about 6-8 years to complete an MIT PhD, Jia Haojun finished in just 5 years. Upon graduating in 2024, numerous large domestic and international companies extended offers, but Jia decided to start his own company. Although almost everyone around him opposed the idea, Jia firmly believed that with the explosion of generative AI and the differentiated advantages achieved by their self-developed ReactiveAI platform in materials discovery, his timing for entrepreneurship was right.

Before securing angel investment from Linear Capital, Jia Haojun hadn't even graduated from MIT and had never met the investors in person. However, their connection was forged during a campus sharing event at MIT.

At that time, Linear Capital founder Wang Huai was invited to share his experience transitioning from an engineer to an investor with students from MIT, Harvard, and other universities. Jia Haojun was among the attendees. In subsequent interactions, Jia left a deep impression on Depth Principle's angel investor and Linear Capital partner Zeng Yingzhe. "Haojun was exceptionally outstanding—a young Chinese man who served as the president of the MIT Chinese Students and Scholars Association at a young age."

During that period, Linear Capital wasn't the only one eyeing Jia Haojun. The two-person team of Jia and Duan, still students with just a几十页 PPT, received dozens of term sheets.

Previously, Jia had planned to accept an investment offer of approximately $3 million from a well-known early-stage incubator fund. However, just before finalization, the other party made last-minute changes to the term details. Jia proactively withdrew from the deal; at that time, he wasn't overly worried about financing.

But to Jia's surprise, capital markets can change rapidly. In 2023, markets entered a contraction phase, and even hot projects often found no interest. Relevant reports showed that the average number of projects invested in and investment scale by institutions dropped by 40% compared to 2022.

In October 2023, Zeng Yingzhe contacted Jia Haojun. "We talked from 8 PM US time until 3 AM, discussing future plans alone for a full 8 hours." After passing this "marathon" due diligence test, Depth Principle finally secured the angel investment from Linear Capital.

Zeng Yingzhe stated that one reason for investing in Jia Haojun was his unique ability to inspire and attract talent.

After Depth Principle was formally established in 2024, it grew from the initial duo to include experienced professionals from renowned companies like Microsoft, Meta, Dow Chemical, BASF, and Saint-Gobain. "Almost everyone joined at a reduced salary," Jia noted, citing the recruitment of COO Zhang Luyang as a representative example.

Zhang Luyang, a former executive at Tenstorrent and Horizon Robotics, had highly attractive offers on the table, including senior management positions and million-RMB salaries. Jia mentioned that Zhang, an internationally recognized expert in "high-performance computing + autonomous driving" technology and products, recognized early on that AI would bring massive changes to science. Since early 2022, he had been deeply involved in Depth Principle's founding as a company advisor.

But being an external advisor is not the same as joining full-time. To persuade Zhang to join, Jia Haojun offered his utmost sincerity. At the time, Zhang had recently become a father, and his family lived in Canada. Persuading him to return to China to start a business required gaining his family's support as well.

To this end, Jia invited Zhang, along with his wife and newborn child, for a trip to West Lake. To let Zhang's wife enjoy the outing worry-free, Jia Haojun pushed the stroller while having a heart-to-heart conversation with Zhang.

"From left to right: Duan Chenru, Jia Haojun, Zhang Luyang."

"He has always been very clear about what he wants and possesses strong execution ability," was the consistent internal evaluation of Jia Haojun at Linear Capital. In March 2024, they reached a $5 million investment agreement. Just over a year later, Depth Principle had completed multiple funding rounds, accumulating hundreds of millions of RMB in total.

During this period, the AI4S赛道 also became a darling of capital. AI for Drug leader Jingtai Technology became the first company to list under Hong Kong's Chapter 18C (Specialist Technology Companies) rules in 2024 and achieved profitability in the first half of 2025. Industry unicorn Shenshi Technology completed a Series C round totaling over 800 million RMB in December 2025, with cumulative financing reaching billions of RMB. Overseas, a16z-led Periodic Labs announced a $300 million funding round in 2025; Lila Sciences secured over $400 million co-led by Flagship and ARK; CuspAI received $100 million from giants like Nvidia to build a materials discovery "search engine."

"I've been sensitive to money since I was young. If you asked me what 15+27 was when I was little, I might not know, but if you asked me what 15 yuan plus 27 yuan was, I could calculate it immediately," Jia Haojun admitted, characterizing himself as a pragmatic person.

Since establishing Depth Principle in Hangzhou, the company has been relentlessly expanding its client base. Jia calculated this clearly: using real projects for training ("using battle for practice") allows them to earn money while improving platform performance simultaneously. In 2025, Depth Principle secured orders worth tens of millions of RMB, serving clients ranging from health and daily chemicals (represented by a European multinational cosmetics giant) to materials and energy (represented by a leading fine chemical plant manufacturer).

Among these, the collaboration with the European cosmetics multinational significantly boosted the company's commercial confidence. In early 2025, this company faced a typical challenge in cosmetics: the stability of active ingredients. In expensive cosmetics, some core molecules are highly active. While they deliver excellent cosmetic effects, this also means they have poor stability and short shelf lives. Therefore, the company wanted to find a ligand additive from over 8,000 molecules listed in China's *Inventory of Existing Cosmetic Ingredients* to enhance the overall formulation's stability.

Traditionally, this work relied on experimental推演, taking months and incurring high costs, with test materials costing up to 10,000 RMB per 50g.

Initially, the company was skeptical about how much value AI could actually deliver. However, persuaded by the Depth Principle team, they signed a Proof of Concept (POC) agreement. Essentially, they gave Jia Haojun a chance to try.

Based on the mechanism of the target molecule's high activity, combined with first-principles and the reasoning capabilities of large models, Depth Principle completed the screening work within a month. The six finally recommended molecules all significantly improved the target molecule's stability, achieving the desired effect.

"After the experiments, the配方性能 and efficiency astonished them," Jia said. A研发 task that typically took years was completed by just two engineers on Depth Principle's platform in a few weeks. This trial run also opened the door for Depth Principle to enter the cosmetics field. The collaboration expanded from the initial POC to deeper strategic cooperation involving molecular design and reaction route optimization.

The joint R&D合作 with this cosmetics giant also helped Depth Principle clarify its commercialization path. Jia坦承 that at the current stage, "collaborating with clients on R&D for specific vertical applications is easier for普及 AI than selling the platform itself." Platformization is laying the groundwork "to cultivate a larger market in the future," rather than being the main revenue source at present.

The managing director from Lenovo Capital suggested that for a startup like Depth Principle to compete with large corporations, it must adhere to a strategy of "taking small steps quickly," first validating the technology's value through small-scale pilots before gradually expanding commercial scale.

In Jia Haojun's plan, as Depth Principle scales up, with the ReactiveAI platform and Agent Mira as the foundation, the company will transition from a "project-based" model to "productization," potentially generating revenue through platform subscriptions (PaaS).

Additionally, a common "last mile" challenge facing the industry is that computationally predicted materials often fail during the synthesis stage. Therefore, in the second half of 2025, Depth Principle began building its own automated laboratory, the AI Materials Factory, to directly address the practical implementation of material synthesis.

In Jia Haojun's view, pioneering in a brand-new field means the business model must be forged through their own efforts, not强行套用 any existing company's template. "AI4S shouldn't stop at the stage of 'selling shovels.' If you believe you have a good shovel, you should go down and dig for 'gold' yourself."

The managing director noted that diverse优秀实践 have emerged within the AI4S industry. For instance, industry leader Shenshi Technology has built a professional drug computation platform and an AI drug chemistry assistant, empowering the entire preclinical drug R&D process and significantly compressing development cycles. Another leading AI4S company, Chengyuan Technology, focuses on the AI + synthetic peptide drug R&D赛道, using its self-developed platform to achieve full-chain AI empowerment across synthetic peptide drug R&D环节, specifically tackling drug development for "undruggable" targets.

However, the director also pointed out that compared to AI4S progress in other fields, chemical materials application scenarios are more fragmented, with numerous small-to-medium batch customization demands, potentially allowing faster commercial落地. Depth Principle holds unique advantages compared to other companies.

In their view, the technical barriers in the AI4S industry are not based on a single环节 but are comprehensive barriers involving "data-algorithm-computing power-interdisciplinary integration." The core barrier is the combination of "high-quality annotated scientific data + domain-specific algorithms."

General-purpose large models cannot meet AI4S demands; domain-specific algorithms must be developed based on the physical and chemical principles of specific disciplines. Depth Principle's ReactiveAI platform is precisely a specialized architecture optimized for chemical reactions and material property prediction. This deep integration of "algorithm + domain knowledge" is difficult to simply replicate.

"Depth Principle's interdisciplinary team and computing power adaptation capabilities are also significant barriers. AI4S requires a composite team of 'AI algorithm experts + materials/chemistry/biology experts + engineering talent.' Building and磨合 such a team is extremely challenging."

The director advised Depth Principle to avoid competing directly in large corporations' areas of strength and instead focus on vertical,细分 scenarios. For example, rather than building a general-purpose AI4S platform, concentrate on niche areas like new energy materials or specialty chemical materials. Binding "technology + scenario" can effectively resist competition from large firms.

However, Jia Haojun's ambition extends beyond competing with domestic players; he aims for the global stage. He believes that in the current AI for Science field, China and the US are starting from the same line, with the industry gathering momentum. At this stage, it's easier to overtake on bends, and Chinese AI4S companies like Depth Principle are poised to turn a new page for the industry.