Decart, an Israel-based developer of real-time world models and ultra-efficient AI infrastructure, has raised $300 million in Series B funding led by Radical Ventures. Its DOS optimization stack powers models including Oasis for interactive world generation and Lucy for real-time video transformation, delivering sub-35ms latency for physical AI and immersive applications. The capital will accelerate scaling of its chip-agnostic inference platform.
Real-Time Demands Reshape Generative AI
The round arrives as demand surges for low-latency inference in robotics, autonomous systems, and live interactive experiences. Decart’s approach emphasizes full-stack optimization across hardware types, contrasting with competitors such as Runway, Luma AI, Stability AI, and Pika Labs that focus more on creative tools or offline generation rather than production-scale real-time physical accuracy.
Latency and Cost Barriers in Current AI
Existing generative systems often struggle with high latency, error accumulation in long sequences, and heavy dependency on specific chips. Decart targets these constraints through its vertically integrated stack that achieves 100x efficiency gains and supports hardware from NVIDIA, Amazon Trainium, and others, enabling persistent interactive environments.
Full-Stack Optimization for Interactive Worlds
Decart built DOS as a hardware-optimized layer for training and inference of world models, paired with Oasis for physics-accurate real-time generation and Lucy for live video editing. This combination supports applications from robotics simulation to streaming platforms, where competitors typically deliver shorter clips without equivalent edge deployment.
"This is what it takes to power Realtime world models - AI systems that respond instantly and evolve with user input."
Strategic Backing Validates Chip Flexibility
Radical Ventures led the round, with participation from NVIDIA as a new investor alongside existing backers including Sequoia. The involvement of a major chipmaker underscores Decart’s value in reducing dependency on single vendors while enabling seamless switching between AI accelerators.
Generative AI Market Expands Rapidly
The broader generative AI market is projected to grow from $83.3 billion in 2026 to $988.4 billion by 2035 at a 31.6% CAGR. Decart positions itself at the intersection of real-time inference and world models, a segment drawing capital as physical AI and edge computing gain momentum. Partnerships with Comcast and NVIDIA for sub-35ms deployments further validate its production readiness.
Product Momentum Continues Post-Raise
Alongside the funding, Decart launched DOS 2.0 and made Lucy 2.0 available via API, targeting 1,600 tokens per second inference and over 100 frames per second world models. The company has maintained profitability since early operations while expanding its team with roles focused on kernel optimization and business development.