Matwings has built a closed-loop capability that combines computational design with experimental validation and industrial deployment. Its work in AI-driven protein design goes beyond algorithms and models.

A piece of plastic can take hundreds, even thousands, of years to break down in nature. In Matwings’ lab, however, waste plastic can be put into a “water” solution and, after ten hours of stirring, broken back down into its original molecules. Those molecules can then be repolymerized into plastic and made into bottles or woven into clothing.

At the World Artificial Intelligence Conference in Shanghai in the summer of 2026, Wu Banghao, CTO of Matwings’ rPET project, explained the process to visitors at the company’s booth. The challenge in recycling plastic, he said, lies in taking it apart. Proteins, meanwhile, are nature’s own molecular scissors, capable of breaking down a wide range of materials.

When people hear the word “protein,” they often think of food. In industry, proteins also include the enzymes in laundry detergent that break down stains and the antibodies used to treat disease. In the past, improving a protein meant lengthy experiments and countless rounds of trial and error. Matwings is now using AI to make proteins into customizable materials for specific applications.

More broadly, replacing some chemical manufacturing with biomanufacturing is seen as a potential force for change across chemicals, pharmaceuticals and materials. AI is helping move that shift from the laboratory to the factory floor.

01 | Beyond the Structure

What companies need is straightforward: proteins that withstand stronger alkaline conditions, enzymes that remain active at higher temperatures, and products with greater activity or yield.

The 2024 Nobel Prize in Chemistry recognized two areas of work: David Baker’s contributions to computational protein design, and Demis Hassabis and John Jumper’s breakthroughs in protein structure prediction. The latter used code to address a problem that had challenged structural biology for decades. The former made designing proteins from scratch a reality.

The academic community was excited by these advances. But Matwings’ founding team saw a different question. “There was a lot of discussion in academia, but when we spoke with companies, we found that many needed more than protein structures. They wanted proteins that could withstand stronger alkaline conditions, enzymes that could tolerate higher temperatures, and functional proteins with greater activity or yield,” Wu said.

AlphaFold is one of the best-known tools for predicting protein structures. It predicts a protein’s three-dimensional structure from its amino acid sequence. Matwings took a different approach: starting with the desired function and designing or modifying the protein sequence to achieve it. This makes it possible to improve existing proteins directly in response to industrial needs.

Matwings describes its development as a three-stage journey. First, its founding team validated the direction through research. Around 2023, the company began delivering projects to external companies through collaborations, building industry experience and credibility. By 2026, with its AI-driven protein development capabilities validated, Matwings shifted its focus toward using its experience in biomanufacturing to build a platform that can support the full research, production and sales cycle across industries. In Wu’s view, the first two stages answered the question, “Can we create better proteins?” The third is what Matwings ultimately wants to do.

One of the hardest parts of protein engineering, Wu said, is bridging the gap between small structural changes and major changes in function. “A protein may have hundreds of amino acids. Changing just one or two may make almost no visible difference to its structure, yet its function can change dramatically.” Functional optimization cannot rely on sequence alone. It requires searching a vast sequence space for a small number of effective combinations. As more positions are changed, the number of possibilities grows explosively, making manual trial and error slow and labor-intensive. This is where AI can help.

One of the collaborations that first brought Matwings wider attention was with GenSci. In biopharmaceutical manufacturing, equipment and media must be cleaned with highly alkaline solutions at pH 13–14. Ordinary antibodies quickly lose activity under these conditions. Using its AI models, Matwings engineered an alkali-resistant single-domain antibody with four times the resistance to extreme alkaline conditions and twice the service life. The antibody was ultimately scaled up to industrial production at 5,000 liters. It became the first reported case of a protein designed by a large model to reach industrial-scale production worldwide. The project showed the industry that an AI-designed protein could be more than a line in a research paper: it could become a product with real economic value.

Business Review learned that Matwings’ collaboration with GenSci has expanded from industrial enzyme engineering to innovative drug development. Wu said that a large-molecule ophthalmic drug has been approved to begin clinical trials for three indications. The drug is intended to treat macular disease, which is relatively common among Asian populations. It can bind to four targets—VEGF-A, VEGF-C, VEGF-D and Ang-2—to jointly inhibit pathological angiogenesis. “You can think of this antibody as a key. In the past, one key opened one lock. This antibody can open four locks at once,” Wu said.

But consistently delivering products like these takes more than one or two success stories. It requires a repeatable production line.

02 | Giving AI a Physical Body

On April 24, 2026, Matwings turned that production line into a conversational product: MatwingsVenus, an AI agent for protein R&D. Users can describe a desired function in natural language. MatwingsVenus breaks the request into tasks, calls on design tools to generate protein sequences, and then connects to an automated laboratory. Robots prepare samples, purify proteins and test their functions. The experimental results feed back into the next round of AI design, creating an intelligent, end-to-end loop of design, validation and iteration.

Wu described the motivation behind the work in four Chinese characters: making science accessible to all. “We have always wanted the benefits of good research to reach more people. We want to take R&D platforms that, in the past, were available only to large pharmaceutical companies and major research institutes and make them accessible to a small team—or even to individuals—as intelligent tools.”

Within just a few months of launch, MatwingsVenus had integrated more than 500 design tools and over 60 AI skills, and brought in more than 50 certified experts to answer users’ questions. Wu said the platform had registered more than 10,000 users, most of them from universities and research teams. “I have many friends who are still in academia. They have no background in computational science, but they can now use MatwingsVenus to carry out design work. They can spend their time thinking about the science instead of struggling to run programs,” he said.

At the heart of MatwingsVenus is the dry–wet loop: “dry” refers to computational design, and “wet” to experimental validation. In biology, experiments remain the gold standard for determining how a protein performs. The efficiency of the loop depends on the accuracy of the AI designs: the more precisely the model identifies promising candidates, the fewer variants need to be tested. Fewer experiments can mean shorter development cycles and lower costs. The design stage is therefore critical, but its value can only be realized through experimental validation.

Wu said Matwings’ underlying design model, VenusREM, has ranked among the leading models on ProteinGym, a global protein engineering benchmark developed by researchers at Harvard Medical School. This year, a Harvard research team compared general-purpose large language models such as GPT and Claude with dozens of specialized models. Matwings’ model ranked first among the specialized models and was used in the study as a leading point of comparison for general-purpose models.

The challenge in running the whole process lies not only in “dry” design, but also in connecting it with “wet” experiments. Automated laboratory equipment has existed for decades, Wu noted. What wet labs have lacked is not machinery, but a “brain” capable of designing experiments and interpreting data. Matwings’ Fully Automated Laboratory is designed to provide one. According to Wu, it can operate without human intervention across the entire workflow: once given a design plan, robots can prepare samples, purify proteins and run tests. The results are then automatically fed back to drive the next round of design.

“This closed loop has brought a fundamental change. The time it takes to move a protein product out of the lab has fallen from two to five years to three to six months. The success rate has risen from around 1% to 30%, and the number of experiments has dropped from tens of thousands to the hundreds,” Wu said.

03 | Entering the Circular Economy

Matwings did not start with a green-economy concept and then go looking for a technology to fit it. It started with its own AI protein design capabilities and found a real industrial bottleneck in the circular economy.

In July 2026, Matwings announced that it had completed a Series A++ round worth several hundred million yuan, co-led by Shanghai State-Owned Capital Investment and SDIC Innovation. Circular economy was listed as the first priority for the funds.

Why would a protein design company put the circular economy first? Wu said the decision was not about simply invoking ESG. Matwings entered the sector where three factors converged: policy direction, industrial demand and the company’s own technical strengths.

At the policy level, the start of the 15th Five-Year Plan period has brought a series of top-level policies on the circular economy and green, low-carbon industry. In June 2026, with State Council approval, the National Development and Reform Commission issued the 15th Five-Year Plan for Circular Economy Development. In July, the Ministry of Industry and Information Technology released the 15th Five-Year Plan for Green and Low-Carbon Industrial Development. The Ecological Environment Code, adopted in March 2026, dedicates a chapter to developing the circular economy. It gives the issue a place in a comprehensive code and strengthens extended producer responsibility, helping move resource recycling upstream—from end-of-life collection toward product design.

The industrial need is even more direct. “Global PET production is close to 100 million tonnes a year, but less than 20% to 30% is actually recycled,” Wu said. “Most of that is mechanical recycling, in which PET is melted and remolded. After two or three cycles, its performance degrades to the point where it can only be discarded. To make plastic truly recyclable, it needs to be broken down at the molecular level—back into its original monomers—and then polymerized again. The main chemical methods today rely on high temperatures and pressures. Enzymatic recycling can break down plastic at ambient temperature and pressure, making it a genuinely green approach to molecular-level recycling.”

The bottleneck in enzymatic recycling lies in the enzymes themselves. Natural enzymes often lack the thermal stability, activity and production yield needed for industrial scale-up. These are precisely the problems Matwings is equipped to address.

Kunlun Capital, a subsidiary of China National Petroleum Corporation (CNPC), was one of the lead investors in Matwings’ previous funding round. The firm said its investment was based on three strategic considerations.

First, the company’s work aligns with CNPC’s green and low-carbon transition and its carbon goals. CNPC is evolving from a traditional oil and gas company into an integrated energy and materials company, with biomanufacturing, the circular economy and green chemicals among its priority areas. Matwings’ work in AI protein design provides an enabling platform technology for synthetic biology and biomanufacturing.

Second, the technology and team have capabilities that are difficult to find elsewhere. Matwings has built a closed-loop capability combining computational design with experimental validation and industrial deployment. That capability can extend into industrial enzymes, bio-based materials and green chemistry, shortening development cycles and reducing the cost of trial and error.

Third, Kunlun Capital sees long-term value in the industrial applications. In its view, AI protein design could bring a paradigm shift to industrial biotechnology, much as AI-assisted drug discovery is reshaping pharmaceutical R&D. Those who can design proteins efficiently are more likely to occupy an upstream position in the future of biomanufacturing.

PET-degrading enzymes are one concrete application of industrial biotechnology. Wu said development proceeded in two stages. In the first, the team used MatwingsVenus to discover a wild-type PET-degrading enzyme with its own intellectual property, completing the work in just over three months. In the second, it spent another three months engineering the molecule to achieve the activity, stability and substrate tolerance required for industrial use—addressing a major enzyme-performance bottleneck in plastic recycling.

04 | Building the Engine for the Age of Biomanufacturing

“We would rather define ourselves as one of the underlying engines of the biomanufacturing era.”

Wu said Matwings is pursuing two business lines in parallel. One is developing protein products—like replacing a chemical process with a greener engine by switching from high-temperature, high-pressure, energy-intensive or low-yield methods to mild and efficient enzyme catalysis. The other is taking large models out of the lab and deploying them locally in downstream factories.

The circular economy is one area where Matwings is putting its technology to work, but it is not the whole story. The company sees itself as an underlying engine for the age of biomanufacturing.

Kunlun Capital has a more specific vision for how that role could translate into collaboration. CNPC has extensive industrial applications across oil and gas development, refining and chemicals. The two companies could work together on industrial enzymes, biocatalysis, bio-based monomers, biodegradable materials and green oilfield chemicals, among other areas. CNPC also has experience in engineering scale-up, supply-chain management and market access, which could help bring Matwings’ technology from the laboratory to industrial production. As an industrial investor, Kunlun Capital can connect Matwings with industrial, customer and policy resources inside and outside the CNPC system, helping create a pathway from technology to pilot scale, industry and market.

Kunlun Capital sees three areas with particularly strong commercial potential for AI protein design. The first is industrial enzymes and biocatalysis, which it views as the most likely to scale in the short term: there is substantial scope to replace existing processes, carbon reductions can be direct, and the economics can be quantified. The second is bio-based and biodegradable material building blocks, including PLA, PHA and bio-based nylon, which combine sustainability with the growth potential of new materials. The third is plastic recycling, particularly enzymatic depolymerization. It is technically challenging, but a breakthrough could create significant industrial and ESG value.

From industrial enzymes to bio-based materials and plastic recycling, these applications point to a shift in the foundations of the chemical industry.

For more than a century, modern chemical manufacturing has been built on high temperatures, high pressures, strong acids and strong bases. This system has given humanity unprecedented productive capacity, but it has also put resources on a one-way path: extract, manufacture, discard.

AI-designed proteins are helping turn that path into a loop. In Wu’s account, the PET plastic that might otherwise take hundreds of years to degrade becomes a resource that can be recovered and used again. As proteins become accessible through natural-language interaction, the way we make things and use resources is beginning to change.