Synopsys CEO says fully autonomous AI chip design remains out of reach

Ghazi's "40 task agents" figure signals a structural shift: EDA's value anchor is migrating from point-solution algorithms to the orchestration layer. Six-dimensional co-optimization across physics, materials, process, architecture, software, and model is a combinatorial constraint-satisfaction problem—no generative model closes that loop without deterministic verification. Technical cascade: the moat is no longer "better layout engine" but "standardized negotiation interfaces." Foundries and IP vendors must expose agent-readable constraint protocols, or 40 agents remain 40 silos. Compliance: once AI agents make design decisions, audit trails become non-negotiable. BIS export controls already inflate multi-region fab compliance costs; layering model-data provenance requirements on top drives expenses up non-linearly. Market positioning: Cadence will likely differentiate on "single-agent depth" to sidestep the orchestration arms race. Design houses in Taiwan, China (MediaTek, Qualcomm) will accelerate internal agent-orchestration stacks to reduce single-vendor lock-in. 12–24 month call: fully autonomous chip design won't ship commercially. But orchestration middleware is the fattest incremental market in EDA right now. Whoever owns the agent-to-agent negotiation protocol owns the next decade of EDA leverage.