How can automakers with existing infotainment systems adopt these advanced AI capabilities without a complete vehicle redesign?

Automakers can use the 'AI box' architecture, which is a modular, standalone computer built on NVIDIA DRIVE AGX. This box connects to the existing infotainment system, typically via Ethernet, and functions as a dedicated AI accelerator. It handles the demanding LLM and VLM processing locally, allowing OEMs to upgrade vehicles with powerful agentic AI capabilities without needing to redesign or requalify their core IVI stack or vehicle electronics.

How to Build In-Vehicle AI Agents with NVIDIA: From Cloud to Car

NVIDIA Outlines Scalable Architecture for In-Vehicle Agentic AI

NVIDIA has detailed a comprehensive framework for building and deploying agentic AI assistants in vehicles, marking a significant step beyond the fixed command-response systems common in today's cars. The proposed architecture addresses the growing demand for conversational assistants that can manage multi-step tasks, understand context, and interact multimodally using voice, vision, and vehicle data. This move provides automakers with a structured path to integrate sophisticated large language models (LLMs) and vision-language models (VLMs) directly into the cabin, a necessary evolution to meet modern consumer expectations for intelligent, adaptive user experiences.

Hardware Pathways for On-Device AI Compute

To address the substantial on-device compute required for real-time AI reasoning, NVIDIA presented several hardware integration strategies built around its DRIVE AGX platform. These solutions are designed to provide the low latency (<500 ms) and high throughput (>30 tokens/sec) needed for fluid conversations while ensuring data privacy through edge-first processing. Automakers can choose the architecture that best fits their needs, from incremental upgrades to full centralization.

AI Box: A modular add-on computer using DRIVE AGX Orin or Thor that augments existing infotainment (IVI) systems. It connects via Ethernet to offload heavy AI workloads, allowing OEMs to add advanced AI to vehicles without redesigning the core IVI stack.
Multi-Domain AI Computer: The DRIVE AGX Thor platform can serve as a unified computer, hosting both autonomous driving and in-vehicle AI workloads on a single, powerful SoC with guaranteed workload isolation.
Central Car Computer: An integrated solution pairing DRIVE AGX with MediaTek's Dimensity AX SoCs. This architecture shares a unified DriveOS software foundation, allowing AI tasks to be offloaded to the dedicated DRIVE AGX chip while the MediaTek SoC handles IVI, gaming, and multimedia.

Hybrid Architecture and Market Impact

The framework extends beyond the vehicle itself, advocating for a hybrid architecture that seamlessly blends edge and cloud AI. This model uses an orchestrator to route tasks to the appropriate agent—local for vehicle control and privacy-sensitive data, and cloud-based for web searches or third-party service integration. This approach is critical for delivering a robust user experience that is both responsive and feature-rich. According to ABI Research, the strategic importance of this technology is clear, with global shipments of vehicles equipped with agentic AI projected to grow from approximately 5 million in 2025 to 70 million by 2035.

Strategic Takeaway: NVIDIA is not just selling chips; it's providing a modular and scalable systems-level blueprint for automakers to integrate complex agentic AI without overhauling their existing electronics architecture, effectively lowering the barrier to entry for next-generation in-cabin experiences.

>> Verify Original Transmission at NVIDIA