Powering the AI Supercycle: Design for AI and AI for Design

At the IEEE International Solid-State Circuits Conference (ISSCC) 2026 in San Francisco, Anirudh Devgan, President and CEO of Cadence, outlined a defining shift for the semiconductor industry:
artificial intelligence is transforming not only the systems we build, but the way we engineer them.

The industry has entered an AI supercycle defined by unprecedented scale and complexity. Demand for intelligence is driving exponential growth in compute, pushing the limits of performance, power efficiency, and system integration. Industry forecasts project the semiconductor market to approach $1.2 trillion by 2030, and current momentum suggests the industry could reach the $1 trillion milestone as early as this year.

But the most important shift is not just what we are building, it is how we design it.

As highlighted at ISSCC, the industry’s next scaling challenge is no longer transistor density alone; it is engineering productivity.

The Engineering Bottleneck of the AI Era

While much of the attention has been focused on the compute requirements of training new frontier models, the growing importance of inference will drive new requirements across semiconductors and the AI infrastructure. These include distributed data centers in closer to population centers to reduce latency, increasing the capability and density of edge compute within energy budgets, and the ever-growing need for high-bandwidth and low-latency connectivity.

These requirements are driving modern systems on chip (SoCs) to integrate hundreds of billions of transistors, advanced packaging, high-bandwidth memory, and heterogeneous compute architectures. Performance is now determined not just at the die level, but across packages, boards, racks, and full systems.

The design space has become too large, too interconnected, and too dynamic for conventional automation alone. Development cycles are compressing from several years to one year or less, even as system complexity accelerates, increasing the cost of late-stage changes and missed power, performance, or area targets.

Meeting the demands of the AI era requires a fundamentally new approach to engineering. Anirudh’s plenary highlights the potential for agentic AI to address these challenges.

From Automation to Agentic Engineering

For decades, electronic design automation (EDA) has delivered productivity gains through deterministic algorithms and point optimizations. But modern chip development is inherently iterative. Engineers explore architectures, refine constraints, evaluate tradeoffs, and repeat this process across multiple domains.

Agentic AI moves engineering from isolated optimization to intelligent exploration.

Instead of operating as standalone tools, AI-powered agents collaborate with engineers to understand design intent, learn from prior runs, and guide decisions toward optimal outcomes. By modeling relationships within high-dimensional design spaces, agentic systems help teams converge faster: reducing rework, shortening schedules, and enabling better architectural choices earlier in the cycle. Early use cases include specification-to-RTL generation, automated verification planning, and AI-driven optimization of power, performance, and area (PPA).

This is not incremental automation. It represents a shift from tool-driven workflows to AI-assisted engineering collaboration.

In many ways, this change marks the beginning of engineering intelligence at scale.

Scaling with Multi-Agent Systems

Modern development spans architecture, implementation, verification, physical optimization, and signoff—each with domain-specific constraints. The next evolution is a coordinated ecosystem of specialized agents working together across the design lifecycle. In this model, engineers define intent and constraints, while multiple domain-aware agents run, iterate, and continuously refine the design state.

This multi-agent approach mirrors how engineering teams already work, now accelerated and scaled by AI that enables exploration with greater speed and consistency.

The Engineering Mental Model

Foundation models alone cannot produce engineering-grade silicon. Semiconductor development requires deep structural understanding, strict constraints, and deterministic behavior.

A key enabler for engineering-grade agentic AI is a structured engineering mental model—a machine-readable representation of design intent that captures hierarchy, interfaces, protocols, connectivity, and system constraints.

By grounding AI generation in this structured knowledge—and combining it with semiconductor-specific data—agentic systems move beyond code generation to true design reasoning. The mental model enables context-aware RTL and verification creation, consistency across the design hierarchy, and traceability from specification to implementation.

This combination of foundation models, domain expertise, and structured design intelligence is emerging as a critical architectural differentiator for scalable agentic EDA.

Early deployments are already demonstrating measurable gains in engineering productivity. By learning from prior designs and organizational data, agentic workflows help teams navigate complex design spaces more efficiently, accelerate convergence, and reduce dependence on manual iteration.

As system complexity increases and time-to-market pressures intensify, faster convergence becomes a strategic advantage. Organizations that can evaluate more architectural options earlier, and reach optimal solutions with greater confidence, will deliver more differentiated silicon.

This engineering transformation is unfolding alongside the rapid expansion of AI infrastructure. Beyond silicon design, AI is also reshaping the infrastructure that powers it. At the system level, digital twins and AI-driven analysis are improving the efficiency and utilization of large-scale AI deployments, with similar model-driven approaches expected to extend into domains such as robotics, autonomous systems, and other forms of physical AI.

But the most immediate competitive differentiation will come from transforming how silicon itself is designed.

Designing the Future, Faster

The semiconductor industry has always advanced through higher levels of abstraction, smarter automation, and deeper integration. Agentic AI represents the next step in that evolution—transforming design from a sequential, tool-centric process into an intelligent, collaborative workflow.

In the AI era, competitive advantage will be defined not only by what companies build, but by how intelligently they design it. The companies that embrace agentic engineering—grounded in domain expertise, structured design intelligence, and human–AI collaboration—will move faster, explore more, and deliver better systems.

In a market accelerating toward the trillion-dollar milestone, the winners will not simply be those who build the most silicon—but those who can design intelligence into it the fastest.

Explore how Cadence is powering the AI Supercycle across the semiconductor design lifecycle at Cadence.com.