Wedoany.com Report, On March 17, 2026, San Francisco-based AI company Anyscale announced new features for Ray and the Anyscale platform, designed to help teams build and deploy AI workloads in production environments. These features target multimodal AI data processing, achieving cost reduction and efficiency gains through the integration of NVIDIA technology.

Anyscale announced the integration of Ray Data with NVIDIA cuDF, enabling GPU-native multimodal data processing and reducing costs by 80%. This feature will be available on AWS EC2 via the NVIDIA RTX PRO 4500 Blackwell Server Edition, helping teams handle complex data such as images, videos, and documents to optimize AI return on investment.

Robert Nishihara, co-founder of Anyscale, stated: "The complexity of AI systems is increasing, from reinforcement learning pipelines that combine simulation, data generation, training, and inference, to multimodal data preparation for RAG and robotics. Ray serves as a unified compute engine for all these GPU-driven workloads, empowering teams with programmatic control to place workloads on the hardware best suited for the task, whether it's the NVIDIA RTX PRO 4500 Blackwell for data preparation or the NVIDIA GB300 NVL72 for large-scale training."

Ray also introduced rack-aware scheduling capabilities for NVIDIA GB300 NVL72 clusters, enabling optimal placement of distributed AI workloads to leverage NVIDIA's high-speed interconnect technology. Developers can express task placement intent through a Python API, and Ray automatically coordinates scheduling to maximize intra-rack bandwidth and reduce cross-rack traffic.

These improvements help advance open-source AI at scale, making multimodal data processing more efficient. Rack-aware scheduling on Ray Data and NVIDIA cuDF support will be available in open-source Ray and the API-compatible Anyscale Runtime, supporting AI teams in accelerating end-to-end experimentation and enhancing production resilience.