Deep Potential Technology's 800 Million Yuan Raise: From Research Workshop to China's DeepMind

"Six Small Stories That Matter"

By: Muxin Xu

Edited by: Zhiyan Chen

Walking down Zhongguancun Avenue through the haze, DP Technology's office hides in plain sight — cold white light and glass curtain walls everywhere, like a Silicon Valley company.

This brings to mind The Thinking Game, the recent DeepMind documentary that took tech circles by storm. Across the ocean, Demis Hassabis and his team work under similar cold lighting, trying to use artificial intelligence to deconstruct the fundamental mysteries of biology.

But a capital storm has also settled here. DP Technology recently completed its Series C funding round, raising over RMB 800 million at a post-money valuation exceeding RMB 6 billion. The round saw continued investment from existing shareholder Beijing AI Industry Investment Fund, with new investors including DCVC, Lenovo Capital, Yuanhe Puhua, and Jingguorui — institutions with deep industrial backgrounds and state-owned capital attributes.

From starting with a few papers in 2018 to now holding hundreds of millions in funding, DP Technology serves as a case study for China's AI4S industry, or perhaps for the commercialization of frontier technology more broadly: when Silicon Valley-style idealism collides with hard commercial reality, how do young people eager to reshape the world with algorithms establish a set of pricing rules in the uncharted territory of science?

Just as a documentary needs chapters to explain the key nodes in a company's development story, DP Technology's seven years can be told through these six small stories.

Part 01

The Alchemist's Fantastic Voyage

Back to DP Technology's starting point. In 2017, on a flight, Linfeng Zhang opened his laptop and used deep learning algorithms to simulate the movement of 64 water molecules over 10 picoseconds. Forty years ago, this experiment would have consumed 200 million core hours; now he needed only a laptop.

Zhang told his Peking University classmate Weijie Sun about this. They both realized that AI had not only calculated accurately but was orders of magnitude faster than traditional molecular dynamics simulation. And this wasn't just a mathematical game — what he held in his hands might be a key to new physical laws governing the microscopic world.

Their first startup capital wasn't venture funding but the top prize from the National Disruptive Technology Innovation Competition: RMB 12 million, paid out over three years. It wasn't until 2020 that this technical intuition was validated by authoritative awards. Zhang, Weinan E, and others won the "Gordon Bell Prize" — the highest honor in international high-performance computing, considered the Nobel Prize of supercomputing — for their breakthroughs in "deep potential" molecular simulation. The achievement was subsequently selected as one of the "Top Ten Scientific and Technological Advances in China in 2020" by the Chinese Academy of Sciences and Chinese Academy of Engineering, alongside the new-generation artificial sun and the quantum computing prototype "Jiuzhang." From then on, DP Technology's fundraising became much smoother.

It was also around this time that Zhang graduated with his PhD from Princeton. According to industry rumors, he once turned down an offer from the DeepMind team to return to China and take charge of DP Technology. He wrote an article announcing this decision, quoting from Paulo Coelho's The Alchemist: "At a certain point in our lives, we lose control of what's happening to us, and our lives become controlled by fate. That's the world's greatest lie."

The next line of that quote corresponds to what his mentor Weinan E told him to encourage his return: the darkest hour of the night comes just before the dawn.

Part 02

Reproducing AlphaFold

Scale and closed-loop operations mean transforming from a "workshop" to a "regular army," which also means attracting top-tier talent. Guolin Ke, now the algorithm partner, joined at this time.

Ke is a technical legend in AI algorithm circles. Before joining DP Technology, he was a senior researcher at Microsoft Research Asia (MSRA) and a core author of the well-known machine learning library LightGBM. The project has over 10,000 stars on GitHub and tens of thousands of paper citations, making it one of the most widely used algorithm tools in global industry.

A small story illustrates his standing in the community — once, while interviewing a candidate, the conversation went well. At the end, the candidate asked for his name. When Ke gave it, the candidate actually stood up and told him seriously: "If I had known you were Guolin Ke, I probably wouldn't have answered your questions so fluently."

In 2021, when Ke first visited DP Technology's office, an intern received him. As they walked down the corridor chatting casually, the intern began introducing DP Technology's biopharmaceutical products and algorithms. Ke was shocked and somewhat excited, thinking: "Even the interns at this company are this impressive?"

That intern later became DP Technology's chief engineer for biopharmaceutical R&D.

Ke decided to join then and there — not just to work with young geniuses, but because he believed AI technology could play a greater role in the AI4S field.

Within weeks of joining, Ke faced what was possibly the algorithm team's most important task: reproducing Google DeepMind's AlphaFold 2.

It was 2021. AlphaFold 2 had just sent shockwaves through the biopharmaceutical field, but DeepMind had not open-sourced its training code. Without a reproducible training implementation, key capabilities were essentially "choked off." So Ke and his team's goal was to achieve training reproduction of AlphaFold 2 with far fewer resources than Google — an extremely perilous campaign. "We almost had no room for trial and error," Ke recalled. Throughout the reproduction process, the core large-scale training experiment was actually run completely only once, and this "one and only time" succeeded.

There was certainly some luck involved, but more critically, the entire team had spent nearly sleepless days and nights repeatedly studying the AlphaFold 2 paper, thoroughly grasping its algorithmic principles, and continuously refining every detail of the engineering implementation.

This experiment later became the world's first complete training reproduction of AlphaFold 2.

Ke then led the team to continue tackling protein modeling challenges. Notably, their protein-small molecule docking model Uni-Mol Docking also attracted DeepMind's attention, and was cited in their subsequent AlphaFold 3 paper as the best-performing baseline model.

Part 03

Not Biotech, but "Microscopic Dassault Systèmes"

Shortly after starting up, DP Technology faced a classic business fork in the road: use algorithms to develop drugs itself (Biotech), or become a software platform company (Platform)?

From a capital returns perspective, the Biotech path seemed more attractive. Developing one blockbuster drug would yield exponential returns. Many AI drug discovery companies indeed chose this route, using AI to screen pipelines and then advancing to clinical trials, trying to become the next Pfizer.

But DP Technology ultimately chose to be a platform.

First, the trust issue. DP Technology's clients are pharmaceutical companies. If DP Technology also developed drugs, it would immediately fall into the awkward position of "being both referee and athlete." Pharmaceutical companies wouldn't feel comfortable handing core data to a potential competitor. Only by strictly maintaining the "no pipeline" red line could DP Technology serve as a neutral third party and early on gain trust from industry giants like Hengrui.

Second, and more importantly, this was about defining the company's vision. DP Technology's ambition lies not in becoming a successful pharmaceutical company, but in becoming "the Dassault Systèmes of the microscopic world" — just as Dassault defined macro-scale industrial design (aircraft, automobiles), DP Technology aims to define R&D standards at the microscopic scale (molecules, atoms).

DP Technology aims to solve the fundamental problem of "atomic arrangement and combination." This capability can be applied not only to pharmaceuticals but also to batteries, semiconductors, and chemical materials. The value of such a general-purpose platform far exceeds the gamble on a single pipeline.

For commercialization, DP Technology chose a "software + services" model. Its first major client was a pharmaceutical company. In the early stages of cooperation, Weijie Sun and the team traveled repeatedly between Beijing, Shanghai, and Lianyungang with demos of the Hermite drug design platform. They didn't pitch grand visions but directly compared data: for a difficult druggable target, did the hit compounds calculated by AI show higher activity and success rates than traditional high-throughput screening?

After lengthy proof-of-concept (POC) validation, this pharmaceutical company ultimately paid. Subsequently, new energy giants like CATL and BYD also became clients, and DP Technology successfully expanded its business from life sciences to materials science, further validating the feasibility of its "microscopic Dassault" platform strategy.

Part 04

The Super Lab Gamble

As commercialization deepened, DP Technology discovered a more fundamental pain point: no matter how accurately AI calculates, if clients can't actually produce the results, it remains "virtual." Moreover, the biggest bottleneck currently facing AI for Science is not algorithms, but the scarcity of high-quality data.

To bridge the gap between "calculating" and "doing," and to solve the data famine, DP Technology made an extremely "heavy" decision: building a super lab (the Bohr·Cyber Lab). It should be explained here that "super lab" is not a room or a production line, but an entire set of experimental infrastructure and scheduling technology systems with Uni-Lab-OS at its core.

One of the lab's heads, Junhan Chang, is a young man born in 1999. He told An Yong Waves that the super lab is not just about freeing hands — its core capability lies in achieving a fully automated closed loop of "reading, calculating, and doing."

"Often, scientists have ideas but get trapped in repetitive physical labor," Chang explained. The super lab operates continuously — once someone submits a request, AI designs the formula, instructions go directly to robotic arms for automatic liquid preparation, reaction, and characterization, with data fed back to the model in real time.

To support the super lab's efficient operation, DP Technology built a rigorous internal division of labor: domain experts "translate" scientific needs into mechanical language; mechanical engineers materialize needs into hardware solutions; and Uni-Lab-OS itself is an open-source laboratory operating system project, supporting both the stable operation of DP Technology's internal super lab and providing external interfaces to research teams, allowing them to build "mini super labs" in their own laboratories as needed.

Moreover, to break traditional research's dependence on "manual experiments," the team needed to go deep into universities and enterprises, using open-source communities, "intelligent experimental equipment" courses jointly offered with universities, workshops, and other means to transform real research topics into reusable automated scenarios, allowing more students and teachers to complete a transformation with their own hands in their own labs.

For the entire AI4S industry, this means finally having a factory that can continuously produce standardized, high-quality "wet lab" data. This data feeds back into AI models, forming a true data flywheel.

Chang's story of joining DP Technology is highly dramatic.

Chang was once a Peking University student. Every summer, Zhang returned to Peking University to hold summer schools. At that time, young Chang was immersed in a theoretical computational chemistry project and took his research ideas to consult with Zhang. Zhang took one look and asserted: "This direction has a ceiling; it won't work for your problem."

Chang wasn't convinced and went back to work on it for another three months. Three months later, he realized it was indeed a dead end. At that moment, he was completely convinced by Zhang's insight into fundamental laws and decided to join DP Technology. Now, this young man born in 1999 is managing DP Technology's "heaviest" asset, trying to solve the "last mile" of scientific experiments.

Part 05

The Signal Behind State Team Entry

Returning to this Series C funding round. The introduction of Beijing AI Industry Investment Fund, Jingguorui, and other state-backed institutions sends a clear signal: AI for Science is no longer merely scientific exploration or a business, but is regarded as national strategic-level scientific research infrastructure. With "chokepoint" technologies frequently emerging, having an independently controllable microscopic R&D foundation carries self-evident strategic value.

However, it cannot be denied that compared to American peers, Chinese AI4S companies still face valuation pressure in capital markets. Sandbox AQ, spun out from Google, raised $500 million in its first funding round, while DeepMind has Google's unlimited financial backing.

Sun has a clear-eyed understanding of this. He believes that in the AI for Science field, there is no significant capability gap between China and the US. In certain aspects, China even has advantages: Chinese scientists have stronger execution capabilities, China has the world's most mature supply chain, and massive amounts of physics and manufacturing data.

So why does capital seem less enthusiastic about Chinese AI4S than American?

"Because we don't have an AlphaFold," Sun said bluntly.

To date, China has not yet produced an AI science star project like AlphaFold that shocks the world and completely "breaks out of the circle." Investors need a super totem, a case that lets them instantly understand the disruptive nature of this technology.

This is precisely DP Technology's ambition for its next phase. Sun expects that in the future, DP Technology will not merely sell software but will create "China's AlphaFold moment" in some vertical field — perhaps a revolutionary solid-state battery material, or a first-in-class RNA drug. Using indisputable results to conquer capital market prejudices.

Part 06

The Shadow of Oppenheimer

At the end of our interview, we discussed the ending of The Thinking Game, where Demis Hassabis mentioned Oppenheimer and the Manhattan Project. When humanity masters god-like tools like atomic energy, fear follows.

This fear, too, DP Technology's founders have experienced. But Sun has more nuanced thoughts.

"The atomic bomb is ultimately just a tool, neutral physical force. But AI is different — an intelligent agent itself is like a 'person,'" Sun said. He fears not only creating uncontrollable tools but also contemplates: if one day, silicon-based intelligent agents truly surpass humans intellectually and even master all laws from the microscopic to the macroscopic, what then is the meaning of human existence?

"Even if that day comes, I think humans still have some unique things that cannot be replaced," Sun said, such as perceptual cognition, such as the experience of art and music.

There is one more thing AI cannot replace: standpoint and choice. The story returns to a basketball court in Fenyang, Shanxi, Zhang's hometown, where one day street "bullies" fought them for the court. Though Zhang resisted fiercely, he still took a beating.

In that moment, Zhang determined a principle that could be called a "creed" — one that has followed him to this day —

"I swear to become a powerful good person."

Image source: Unsplash

Recommended Reading

The flow of money, the rise and fall of people