The Infrastructure Powering Live AI | Decart x Bloomberg | RAISE Summit 2026

Company Positioning & Capabilities

• CART (The CART) is a foundational AI lab with approximately 100 experts focused on building the fastest AI stack, ranging from custom assembly code for GPUs to proprietary model training.
• The company pioneered "real-time world models" specifically for physical AI applications, including autonomous vehicles, drones, and humanoids.
• They claim to be the only company globally (West and East) possessing a live video generation model.
• Their core optimization layer, "DOS" (The CART Optimized Stack), runs AI models one to two orders of magnitude faster than competitors.

Product Launches & Partnerships

• "Lucy" Model: A live AI interface enabling photorealistic virtual try-ons for clothing and furniture in real-time video feeds.
• E-commerce Partners: Working with Amazon and eBay (an investor) to enable users to overlay items seen on e-commerce sites directly into live camera feeds.
• Performance Metrics: Partners report that using this technology has doubled or tripled sales volumes on peak trading days.
• Gaming & Ads: Significant traction noted in gaming and advertising sectors over the past quarter.

Key Use Cases & Customer Adoption

• Live Streaming: Top-tier streamers on Twitch, TikTok, and YouTube are using the tech to increase engagement; for example, streamer Code Miko saw seven times her average engagement during streams featuring the technology.
• Interactive Shopping: Viewers can place objects from their own homes into live streams, creating new dynamic interactions between sellers and buyers.
• Enterprise Deployment Strategy: CART deploys forward-deployed engineering teams to partner sites (e.g., San Francisco, New York) for 6–12 month residencies to customize integration and drive specific business KPIs.
• Coding Agents: Emerging application involving "5 to 10x faster coding agents," leveraging the high-performance stack to solve software engineering tasks more efficiently.

Technical Architecture & Training

• Training Data: Models are trained on text, images, and "millions of hours of video" to simulate physical properties like fabric dynamics, physics, and structural integrity.
• Hardware Collaboration: Deep integration with chip manufacturers to extract maximum performance from hardware accelerators, co-creating the tech stack for real-time generation.
• Data Philosophy: To achieve high fidelity in specific domains (e.g., swimsuits), the AI requires exposure to a vast range of real-world scenarios (e.g., pool jumping) to generalize physics correctly.

Future Roadmap & Market Outlook

• Robotics Timeline: Autonomous vehicles are expected to see drastic takeoff in 18–24 months; humanoids and drones are similarly projected within the same window.
• Robot Simulation Strategy: The technology enables robots to "imagine" the consequences of actions in a simulation before executing them in the physical world.
• Software Engineering: Predicts solving "lots of software engineering" and dropping the cost of translating imagination to reality to zero within 12–18 months.
• Market Philosophy: Advocates for rapid experimentation to identify "10 compelling use cases" rather than dozens of minor variations, noting that user behavior evolves quickly (e.g., chat transforming internet interaction).