Key Strategic Insights

1. Edge Inference Becomes the Default for Autonomy

Cloud-based inference is no longer viable for most robotics applications. Latency, bandwidth and safety requirements dictate that intelligence must move directly onto the robot.

What this requires

• Compute platforms with integrated NPUs
• Robust AI pipelines optimised for low-latency execution
• Reliable local inference even under power constraints

2. Heterogeneous Compute Architectures Take Over

Robots increasingly rely on multiple compute domains working together: CPUs for orchestration, GPUs for vision, NPUs for inference and FPGAs for deterministic operations.

What this requires

• Platforms designed for multi-accelerator integration
• Unified toolchains for CPU/GPU/NPU pipelines
• Scalable module families (SMARC, OSM, COM-HPC) to keep flexibility[KP1]
• Motherboard or box PC platforms enabling plug-and-play integration, with ready-to-connect sensor interfaces, pre-flashed OS and application environments, and existing certifications to accelerate deployment.

3. Sustained TOPS/Watt Becomes the True AI Performance Metric

Peak performance values say little about real-world robotics workloads. The ability to deliver stable throughput over long operation cycles – without thermal throttling – is becoming the defining characteristic of AI compute.

What this requires

• AI SoCs optimised for efficiency, not just peak TOPS
• Better power budgeting and thermal profiling
• Stable inference under continuous load

4. Silicon Flexibility Becomes a Strategic Necessity

Compute lifecycles are getting shorter, while robots remain in the field for years. Being locked into a single vendor or SoC generation is now a major strategic risk.

What this requires

• Standardised module formats for vendor interchangeability
• Carrier boards designed around SMARC, OSM or COM-HPC pinouts
• Future-ready platform planning to avoid forced redesigns

5. Certification & Compliance Drive Demand for Pre-Validated Platforms

With safety standards increasing across industrial and collaborative robotics, compute modules that come with known EMC behaviour and documented compliance paths reduce significant risk.

What this requires

• Platforms with established certification baselines
• Reference designs and certified box PC platforms aligned with IEC/ISO safety standards
• Reduced integration and documentation overhead

[KP1]4th bullet:

Motherboard or Box PC for plug and play solution to connect the sensors, flash the OS and application all already certified.