Pathway announced the implementation of its post-Transformer BDH (Dragon Hatchling) architecture on NVIDIA AI infrastructure and AWS’s cloud and AI tech stack, enabling a new class of adaptive and continuously learning AI systems. This initiative leverages a brain-like architecture, challenging conventional deep learning assumptions and unlocking continuous learning capabilities through dynamic state management and Hebbian synaptic plasticity. According to Zuzanna Stamirowska, CEO and co-founder of Pathway, this integration represents a fundamental shift from static to adaptive intelligence, opening new possibilities for complex applications previously unattainable for enterprise customers.
Pathway’s BDH Architecture and Capabilities
Pathway’s groundbreaking BDH (Dragon Hatchling) architecture represents a shift from static to adaptive intelligence, challenging conventional deep learning assumptions. Unlike traditional Transformer-based models with consistent knowledge states, BDH utilizes a biologically inspired, scale-free network design. This allows for continuous learning via dynamic state management and Hebbian synaptic plasticity, enabling models to evolve alongside business operations. The architecture is designed for enterprises needing complex thinking, low latency, or full observability into a model’s live state.
BDH aims to improve upon existing AI models by demonstrating that scale can increase interpretability through neuron specialization—a departure from the belief that larger models become less understandable. This architecture achieves competitive performance on general-purpose hardware and shows potential for faster inference on specialized AI processors, offering opportunities to reduce latency and operational costs. Pathway’s model outperforms Transformers and provides enterprises with full visibility into how the model works.
The BDH architecture is now running on NVIDIA AI infrastructure and AWS’s cloud, with availability through AWS beginning in late 2025. Pathway is collaborating with AWS as its preferred cloud provider to deploy the adaptive architecture at enterprise scale, leveraging the NVIDIA Hopper architecture for high-performance, low-latency, and continuous model adaptation. The architecture will be showcased at AWS re:Invent in Las Vegas, December 1-5, 2025.
NVIDIA and AWS Technology Integration
Pathway’s groundbreaking BDH (Dragon Hatchling) architecture now operates on both NVIDIA AI infrastructure and AWS’s cloud and AI technologies. This integration represents a shift from static to adaptive intelligence, unlocking complex applications previously unattainable for enterprise customers. The collaboration leverages AWS as Pathway’s preferred cloud provider, supplying the necessary compute infrastructure for BDH’s adaptive architecture at scale, while NVIDIA’s Hopper architecture provides an optimal environment for high-performance, low-latency, and continuous model adaptation.
Pathway’s BDH architecture challenges conventional deep learning assumptions, suggesting that scale can increase interpretability through neuron specialization. Unlike traditional Transformer-based models that maintain consistent knowledge, BDH utilizes a biologically-inspired, scale-free network design, enabling continuous learning via dynamic state management and Hebbian synaptic plasticity. This allows models to evolve with business operations, offering advantages for applications needing adaptive intelligence over time, and potentially reducing latency and operational costs.
The BDH model is available through AWS, with initial design partnerships slated to launch in late 2025. Pathway intends to showcase the architecture at AWS re:Invent in Las Vegas, December 1-5, 2025. The company highlights that BDH achieves competitive performance on general-purpose hardware, while demonstrating potential for significantly faster inference on specialized AI processors—benefitting enterprise deployments. Pathway is focused on delivering AI that evolves alongside business needs, contrasting with the static nature of many current models.
Enabling innovators to bring new technologies to market is exactly what AWS was created to do.
Jason Bennett, VP and Global Head of Startups and Venture Capital at AWS
Shifting from Static to Adaptive Intelligence
Pathway is delivering a shift from static to adaptive intelligence with its new BDH (Dragon Hatchling) architecture, a post-Transformer model. Traditional Transformer-based models deliver consistent knowledge, but BDH is designed for continuous learning, evolving with business operations rather than remaining fixed. This is achieved through a biologically-inspired, scale-free network and dynamic state management, allowing the model to adjust internal representations as new evidence arrives – unlocking applications needing adaptive intelligence over time.
The BDH architecture challenges the assumption that larger models become less interpretable; Pathway suggests scale can increase clarity through neuron specialization. Utilizing Hebbian synaptic plasticity, BDH offers a framework where scale and interpretability grow together, offering full observability into how the model works. This contrasts with conventional deep learning approaches and is designed for applications requiring complex thinking, low latency, or high observability.
Pathway’s adaptive BDH is being deployed on NVIDIA AI infrastructure and AWS’s cloud, providing the compute necessary for enterprise scale. The collaboration leverages the NVIDIA Hopper architecture for high-performance, low latency and continuous model adaptation. Initial design partnerships are launching in late 2025, and the architecture will be showcased at AWS re:Invent in Las Vegas, December 1-5, 2025.
