First rack-mountable, safety-capable infrastructure, designed for robotics, medical, and security workloads.

The convergence of AI and physical systems is moving fast, and it’s moving out of the lab. From machines that interpret vision and sensor data in real time to systems operating under strict safety constraints, physical AI requires edge infrastructure designed for industrial-grade computing, rather than typical datacenter servers.

That’s the design goal behind Inventec’s Edge IGX 2U Atlas Server, built on NVIDIA IGX Thor platform: NVIDIA IGX T7000 boardkit. It’s a rack-mountable edge system engineered for deployment environments where performance matters, but reliability, manageability, and safety capabilities matter just as much.

A rack-mountable edge platform designed for enterprise operations:

Edge AI is increasingly deployed in places that still need data center discipline: predictable thermal behavior, standard power distribution, remote management, and scalable rollouts. The Edge IGX 2U Atlas Server uses a 2U rack-mountable form factor to fit into existing enterprise infrastructure, enabling consistent deployment alongside standard rack cooling, power, and management practices.

To scale edge deployments responsibly, visibility is not optional. The system includes a Base Management Controller (BMC) that provides continuous telemetry and monitoring without consuming application workload resources. For high-throughput networking, it integrates NVIDIA ConnectX-7 SmartNIC with 2x200 GbE connectivity, supporting data-heavy sensor and video pipelines common in physical AI environments.

Configurable acceleration for edge workloads:

Physical AI workloads can vary significantly in compute profile, from sensor ingestion and perception pipelines to advanced AI inference, including vision-language models. The Edge IGX 2U Atlas Server supports optional NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition or NVIDIA RTX PRO 5000 Blackwell GPUs, enabling configurations that scale based on workload intensity and deployment constraints. With iGPU plus dGPU, the system delivers up to 5,581 FP4-Sparse TFLOPs of AI compute, providing headroom for demanding real-time inference at the edge.

Platform Foundation: NVIDIA IGX Thor and the Software Ecosystem:

At the core of the platform is NVIDIA IGX Thor, engineered for mission-critical industrial and medical edge applications. It includes 96GB of high-speed dGPU memory and supports up to four Multi-Instance GPU (MIG) instances. MIG enables teams to partition GPU resources into isolated slices, allowing multiple workloads to run simultaneously with reduced resource contention. This is useful when you need parallel pipelines, for example perception and sensor fusion alongside model inference and orchestration, without one workload starving another.

Performance alone doesn’t make a platform enterprise-ready. NVIDIA IGX Thor is designed for resilience, with industrial-grade specifications such as extended temperature and vibration ratings and a 10-year hardware lifecycle. It also includes on-chip safety extensions and an integrated safety microcontroller to support developers building systems that must meet functional safety requirements.

Edge deployments succeed or fail on the software runway. NVIDIA IGX Software 2.0 provides a development ecosystem aimed at high-performance sensor processing and safety-oriented AI. It includes frameworks such as NVIDIA Holoscan for real-time sensor pipelines, along with edge AI tooling including the Edge LLM SDK and NVIDIA NIM microservices that support production deployment patterns.

For enterprises planning multi-year programs, lifecycle support is critical. NVIDIA AI Enterprise provides production-grade software support, ensuring teams have continuous access to optimized frameworks, regular updates, and curated models throughout long-term deployments. 

From early engagements to production readiness:

Inventec is already working with select partners and customers through early engagements, with multiple physical AI use cases under development across robotics, healthcare, and security domains. These collaborations inform system configurations, software integration strategies, and deployment architectures ahead of broader rollouts.

Furthermore, Inventec moves from platform details into practical scenarios, showing how this infrastructure supports real-world robotics development workflows, AI-assisted medical imaging, and mission-critical security operations.