An ASIC is a system, but it also serves as a subsystem within larger systems, such as IoT devices or smartphones. System analysis begins with defining use cases and establishing functional and performance requirements. The result is a system specification that outlines key components along with critical metrics such as cost, performance, power consumption, and area.

Choosing the right ASIC is critical when developing products. Beyond business and sourcing strategies, technology node selection, key evaluation metrics include cost, performance, power efficiency, and size. For new ASIC development, a system-based approach—integrating technology node selection, foundry choice, architecture design, IP blocks, and HW/SW interfaces while tracking key metrics—ensures efficient and cost-effective execution. 

Developing an AI strategy begins with a deep understanding of both internal and customer-facing use cases. Training data is critical, requiring a well-defined acquisition strategy—whether sourced internally, externally, or through synthetic generation.

Next, key architectural decisions must be made: Should AI be cloud-hosted or edge-based? Should the model be custom-built or leverage existing commercial models? If using commercial models, techniques like fine-tuning, prompt engineering, and retrieval-augmented generation (RAG) must be explored to tailor performance.

AI models rarely perform optimally without rigorous optimization, leveraging both known techniques and novel approaches. The deployment model must also be carefully chosen—whether AI agents, Autonomous AI, AI Assistants, or AI plugins—to align with business objectives.

Equally important is AI/ML Ops, ensuring efficient development, monitoring, and maintenance of models in production. Automation and continuous improvement pipelines must be established to sustain long-term performance.

Finally, before execution, a clear return on investment (ROI) must be defined, ensuring that AI initiatives drive meaningful value, align with business goals and avoid getting caught in a dotcom kind of bubble.