Aureka Biotechnologies announced the completion of a US$100 million Series B financing round on August 10, 2026. Granite Asia provided the first tranche as the sole investor, while a leading strategic investor led a subsequent tranche that included participation from HighLight Capital (HLC) and additional investments from existing shareholders such as MPCi and NRL Capital. The latest financing brings Aureka’s total capital raised to nearly US$200 million.

The company plans to allocate the new capital primarily to research and large-scale training of its next generation of biological foundation models. These efforts are intended to enhance performance in areas including de novo molecular design, biological structure modeling and biological function prediction. Aureka will also expand Lab-in-the-Loop, its experiment-driven feedback platform, to create a stronger connection between its foundation models and proprietary single-cell functional screening, high-throughput experimental validation and drug development capabilities.

Having established an AI-native, closed-loop R&D infrastructure, Aureka is now concentrating on strengthening the intelligence layer at the center of that infrastructure: its foundation models. The company combines large-scale pre-training, program-specific post-training, AI agents and high-throughput experimentation to create an AI-for-Science platform for life sciences.

Rather than limiting AI to individual drug discovery tasks, Aureka aims to develop models capable of learning biological principles and using that understanding to interpret, generate, predict and ultimately influence complex biological systems.

As foundation models increasingly merge with automated R&D, Aureka is pursuing a broader ambition: moving beyond the use of AI as a tool for improving individual stages of drug discovery toward AI systems capable of modeling living systems themselves. The shift could expand both the technological capabilities and commercial opportunities of AI-powered drug discovery.

Closed-Loop AI Infrastructure Strengthens the Model Intelligence Layer
Established in 2023, Aureka Biotechnologies is an AI-native TechBio company focused on biological foundation models and the closed-loop infrastructure supporting them. Its platform brings together AI models, autonomous agents, digital biology and experimental technologies with the goal of transforming the drug discovery process from end to end.

Biological discovery cannot be driven by computation alone. Models must continually receive information from real-world experiments. Improving large biological models therefore requires more than additional computing resources, better algorithms or new model architectures. It also requires reliable experimental data that captures molecular behavior and an experimental environment capable of repeatedly testing hypotheses, identifying errors and incorporating new findings into subsequent model iterations.

Aureka has positioned Lab-in-the-Loop as a central component of this development process. The system combines AI agents, high-throughput digital biology, proprietary single-cell functional screening and internal experimental capabilities. This creates an integrated workflow spanning molecular generation, experimental planning, functional testing, model post-training and candidate development.

Under this approach, the laboratory is not simply used to verify computational predictions after they have been produced. Instead, experimentation becomes part of the learning process itself. Models generate molecular designs and scientific hypotheses that can be experimentally tested. The company's experimental systems then produce functional data, which is fed back into both its foundation models and program-specific models. The resulting information supports another cycle of design, testing and improvement.

This approach also allows Aureka to produce large volumes of information-rich functional experimental data that can support foundation-model pre-training, reinforcement learning and program-specific post-training. Unlike approaches built predominantly around publicly available and static datasets, Aureka's models can receive continuous experimental feedback from active drug discovery programs.

The result is a recurring cycle in which data, models, experiments and emerging drug candidates continuously strengthen one another.

Independent Evaluation Supports Foundation Model Performance
Aureka's infrastructure has produced AuraIDE, the company's biological foundation model. Trained extensively using proprietary protein co-evolution data, the model is designed to capture relationships among protein sequence, structure, evolutionary history and biological function. The company reports that AuraIDE has achieved leading performance in areas including biomolecular structure prediction and de novo molecular design.

AuraIDE is designed as a general biological foundation model rather than a tool dedicated to a single application. Through task adaptation and program-specific post-training, its capabilities can be transferred across different drug discovery initiatives. These capabilities include protein structure prediction, molecular generation, biomolecular interaction modeling, functional prediction and optimization involving multiple design objectives and constraints.

Aureka's open-source model, OpenDDE, has also performed strongly in independent third-party assessments, ranking among the leading open-source biomolecular models evaluated to date.

Together, independent assessments and experimental results suggest that Aureka's models can perform strongly in protein structure prediction and de novo molecular design while also translating computational predictions into measurable biological activity. Through repeated Lab-in-the-Loop feedback, the company is working to move beyond predicting biological structures toward designing biomolecules with specific intended functions.

Commercial Applications Convert Model Intelligence into Drug Assets
Aureka is combining its biological foundation models, proprietary single-cell functional screening technology and program-specific post-training to develop differentiated antibodies at scale. The company is targeting areas that can be difficult for conventional discovery methods, including challenging targets such as G protein-coupled receptors (GPCRs) and dual-target antibodies designed to interact with two targets through a single molecule.

For individual programs, Aureka adapts its foundation models to the specific biological mechanism, functional characteristics and developability requirements associated with the target. This process transforms broad biological knowledge into a specialized model tailored to a particular discovery challenge.

AI agents can then coordinate activities spanning target analysis, molecular generation, computational evaluation, experimental planning and interpretation of experimental results. Data generated during those experiments is subsequently incorporated into the relevant program model, creating another cycle of refinement.

Aureka's end-to-end agentic R&D infrastructure links molecular generation with developability analysis, experimental testing, feedback and candidate advancement. By connecting these capabilities, the company aims to accelerate the transition from scientific hypotheses and computational models to drug candidates suitable for development.

The platform supports both Aureka's internal pipeline and collaborations with external partners. The company has established strategic relationships with several major global pharmaceutical companies to advance differentiated antibody therapeutics. It also reports generating tens of millions of dollars in revenue during the past two years, which it views as evidence of the platform's ability to deliver at scale in real-world drug discovery programs.

Moving Beyond Efficiency Toward a Biological World Model
“When advanced biological foundation models are combined with R&D infrastructure capable of operating at scale, the objective changes from optimizing a single stage of drug discovery to creating a new generation of discovery engines capable of understanding, generating and predicting biological systems,” said Dr. Weian Zhao, Founder and Chief Executive Officer of Aureka Biotechnologies. “This represents an important milestone in our longer-term effort to develop a biological world model.”

Aureka's longer-term vision extends beyond predicting static molecular structures. The company aims to develop systems that can model molecular interactions, anticipate the consequences of molecular engineering and design decisions, and enable AI agents to autonomously plan, execute and refine drug discovery workflows.

The latest financing will support further development of the relationship between Aureka's biological foundation models and its closed-loop AI-native infrastructure. The company intends to accelerate the testing and real-world application of its models in active drug discovery programs while expanding the role of generative AI in antibody development.
Investor Perspectives

Granite Asia — Yinghui Kuang
Granite Asia views AI-driven drug discovery as entering a stage in which progress will depend increasingly on the interaction among data, models and experiments rather than on isolated improvements in model performance. The firm believes Aureka's AI-native R&D infrastructure gives it the ability to continuously generate high-quality experimental data, improve its models and translate those capabilities into drug candidates. Granite Asia is optimistic about Aureka's long-term potential in generative antibody design and its prospects for international expansion.

HighLight Capital (HLC)
HighLight Capital said Aureka's integration of generative AI with high-throughput wet-lab capabilities could contribute to a significant change in biologics R&D. The firm highlighted the team's technical expertise in AI-enabled drug development and the efficiency of its closed-loop approach. HLC also expressed its intention to continue supporting Aureka's pipeline development, technological advancement and international application of AI in drug innovation.

MPCi — Yuye Wang
MPCi described Aureka as a company it supported at an early stage and continues to have confidence in. It highlighted OpenDDE, Aureka's open-source all-atom model, as evidence of the team's technical depth and early recognition of AI's potential in drug discovery. MPCi also pointed to Aureka's differentiated pipeline and expressed confidence that the company's combination of technical innovation and strategic execution can overcome persistent challenges in drug development and contribute to broader industry transformation.

NRL Capital
NRL Capital said Aureka is helping establish a new infrastructure model for AI-powered drug discovery. The firm views the open-source release of OpenDDE as an important step toward building a broader platform that connects computational models with highly automated, high-throughput laboratory experimentation.

NRL Capital also emphasized Aureka's ambition to establish a globally accessible ecosystem for AI-native drug discovery and praised Dr. Weian Zhao and his team for pursuing an open-source approach. The investor believes that tightly integrating computational and experimental workflows can deliver substantial improvements in antibody design efficiency. NRL Capital has made an additional investment in the latest financing round and plans to continue supporting Aureka's international expansion and efforts to realize the broader value of its AI infrastructure.

A growing school of thought within artificial intelligence suggests that the long-term competitive advantage in AI may no longer reside primarily in the models themselves. As advanced AI systems increasingly achieve similar capabilities, many experts believe that the true source of differentiation is shifting toward the proprietary, structured data that these models use for reasoning and decision-making. Operating from this perspective,

MindWalk Holdings Corp., a company focused on Bio-Native AI, has submitted a European patent application aimed at protecting the high-dimensional biological data structures that underpin its HYFT platform.

Key Highlights

Moving Beyond the AI Model
The central premise behind MindWalk's patent strategy is that AI models themselves are becoming increasingly interchangeable. As leading models continue to converge in capability, the company believes that enduring value will come from proprietary biological context and structured knowledge representations rather than from the models alone.

In June 2026, the Austin-based company announced the filing of European patent application EP26187897.9. The application targets high-dimensional representations of biological subsequences and associated property inference methodologies. Specifically, the filing aims to protect the enriched biological architecture that supports HYFT Technology, the ReefIQ biological context layer, and the LensAI reasoning environment.

According to Jennifer Bath, Ph.D., President and Chief Executive Officer of MindWalk, the long-term question in AI is not which model is being used, but rather the quality and structure of the biological information upon which the model operates. She argues that within life sciences, the differentiating factor is the underlying biological representation system that enables AI models and autonomous workflows to retrieve connected evidence, preserve provenance, and leverage accumulated knowledge across multiple research programs.

MindWalk positions its filing against a broader trend emerging in scientific AI: powerful models alone are insufficient for solving complex biological problems. The company points to publicly disclosed initiatives such as NVIDIA's BioNeMo Agent Toolkit and AstraZeneca's ChatInvent platform as examples demonstrating the importance of domain-specific knowledge, structured interfaces, provenance tracking, memory systems, and validation mechanisms in scientific AI applications.

Extending the Existing Foundation
The newly filed patent builds upon MindWalk's foundational HYFT patent (WO 2020/161344), which established a methodology for identifying recurring biological patterns across living systems and using those patterns as a searchable language for sequence comparison without traditional alignment methods.

MindWalk states that the new application protects a separate and complementary computational layer that organizes biological meaning around those recurring patterns. This layer is intended to enable reuse across the company's internal systems, customer programs, and AI-driven workflows. Rather than replacing the original patent, the company describes the new filing as protecting an additional architectural component built atop the existing foundation.

The distinction between this approach and purely model-centric AI systems forms a key part of MindWalk's thesis. While large language models can capture extensive knowledge, much of that information remains embedded within model parameters, making it difficult to inspect, update, or govern in regulated scientific environments.

MindWalk's architecture seeks to address this challenge by maintaining a biology-aware representation layer that connects meaningful biological patterns with associated sequence information, structural characteristics, physicochemical properties, functional annotations, experimental results, and literature-derived evidence. This information can then be retrieved, updated, compared, and reused as scientific knowledge evolves, without requiring complete retraining of underlying AI models.

Addressing Fragmented Biological Data
One of the persistent challenges in pharmaceutical discovery is the fragmentation of scientific information. A single research program may generate sequence data, structural analyses, physicochemical measurements, experimental results, literature references, and historical decision records that become distributed across numerous databases, teams, and software environments.

MindWalk argues that such fragmentation causes both researchers and AI systems to lose valuable contextual relationships. The company's proposed architecture is designed to preserve those relationships by maintaining links between biologically meaningful patterns and the contextual information explaining their significance.

According to Dirk Van Hyfte, M.D., Ph.D., Chief Technology Officer of MindWalk, biological understanding cannot be isolated into a single data format. Instead, sequence information, structure, function, physicochemical behavior, supporting evidence, and scientific literature must remain interconnected if AI systems are to generate meaningful insights. The company states that its patent filing aims to protect precisely this organizational framework.

Applying the Architecture to Research Programs
MindWalk reports that it has begun applying its approach within active research programs, although all results disclosed to date remain preclinical.

In dengue research, the company has reported binding-level preclinical data showing that targets identified through HYFT informed immunogen design efforts that produced antibodies capable of binding antigens from all four dengue virus serotypes across two separate studies.

Similarly, in influenza research, MindWalk has identified a functional constraint through HYFT analysis that appears across extensive influenza A and B datasets, including human, avian, swine-associated, Victoria, and Yamagata strains.

The company emphasizes that these findings remain preliminary and that substantial additional work will be required to evaluate factors such as neutralization efficacy, safety, durability, regulatory feasibility, clinical translation, and commercial viability.

This research strategy reflects what MindWalk describes as its functional and evolutionary constraint hypothesis: the idea that recurring biological patterns persist because they serve important roles related to structure, function, binding interactions, immune recognition, or evolutionary fitness. By preserving both the patterns and their surrounding context, the company aims to provide AI systems with a more transparent and biologically grounded reasoning framework.

Commercial Implications and Investor Perspective
MindWalk's commercial implementation of this strategy is embodied in its ReefIQ and LensAI platforms. The company reports that LensAI currently operates under recurring commercial agreements with life sciences customers and that the patent filing seeks to protect the foundational layer supporting those deployments as biological data and customer experience continue to accumulate.

Within the company's architecture, HYFT identifies biologically meaningful pattern anchors, ReefIQ organizes biological and customer data around those anchors within a governed context layer, and LensAI performs reasoning tasks that support target identification, candidate evaluation, hypothesis generation, and portfolio decision-making.

MindWalk believes this approach addresses a rapidly expanding market opportunity. Based on third-party industry projections cited by the company, spending on AI technologies for drug discovery could grow from approximately US$5 billion in 2026 to more than US$8 billion by 2030, complementing the pharmaceutical industry's annual research and development expenditures exceeding US$250 billion. The company notes that these figures represent external forecasts and are subject to uncertainty.

From an investment perspective, MindWalk presents the patent filing as part of a broader strategy to build value independent of any individual AI model. The company argues that its biology-aware representation layer constitutes a model-agnostic infrastructure asset whose value may increase as additional programs, datasets, and customer relationships become integrated into the system.

Broader Industry Context
MindWalk positions itself as a Bio-Native AI infrastructure company and emphasizes that comparisons with other public companies serve only as industry context.

Absci represents an approach centered on combining generative AI with synthetic biology and high-throughput laboratory validation for antibody discovery.

Certara operates within the biosimulation and model-informed drug development software market, providing a perspective on the established software infrastructure supporting pharmaceutical research.

AstraZeneca exemplifies the pharmaceutical industry's adoption of agentic AI systems within real-world discovery environments, including initiatives such as ChatInvent.

NVIDIA supplies much of the computational infrastructure and software ecosystem that powers contemporary AI applications, including tools designed specifically for life sciences research.

While these companies occupy different positions within the ecosystem, together they illustrate the breadth of technological approaches shaping AI-enabled drug discovery.

Conclusion
Filing a patent application represents the beginning of a process rather than a guarantee of protection. European patent examination may ultimately narrow, modify, or reject claims, and the eventual scope, enforceability, and commercial value of any granted patent remain uncertain. MindWalk itself acknowledges these risks, as well as the early-stage nature of its dengue and influenza programs.

Nevertheless, the company's strategic thesis remains clear: as AI models become increasingly commoditized, lasting competitive advantage in life sciences AI may derive from the structured biological knowledge systems that support those models. Through this filing, MindWalk is seeking to secure intellectual property protection around its own interpretation of that foundational layer.

For investors interested in identifying where durable value creation may occur as the AI ecosystem evolves, MindWalk's patent filing provides a noteworthy indicator. The ultimate significance of this strategy will likely depend on future patent outcomes, commercial adoption, and the company's ability to generate sustained revenue growth.