Look at the numbers. In 2024, the global semiconductor market hit $600 billion. EDA tools—the software that designs every modern chip—accounted for less than 3% of that, roughly $15-18 billion. Yet for every dollar of EDA revenue, the industry generates $200-300 of semiconductor output and $5,000-10,000 of end-user tech value. That is a leverage ratio market has not priced in. Cadence, the second-largest EDA player, sits at the center of this asymmetry. Its CEO recently argued the company is undervalued amid the AI boom. The data suggests he is not wrong—but the real story is about a structural shift in how EDA captures value, not just a cyclical AI tailwind.
Context: Why EDA Is the Invisible Axe in the AI Gold Rush
Cadence’s core business is Electronic Design Automation (EDA)—the software used to design, simulate, and verify integrated circuits. Without it, no chip—AI GPU, ASIC, or otherwise—leaves the drawing board. The company also sells intellectual property (IP) blocks (PCIe, DDR, SerDes, etc.) and system-level analysis tools. Together, these form a full-stack platform that covers every step from RTL to GDSII, supporting nodes from 28nm down to 2nm GAA.
Cadence operates in a duopoly with Synopsys, controlling roughly 30-31% of the global EDA market. Its customer list reads like a who’s who of chip design: Nvidia, Apple, AMD, Qualcomm, Broadcom, and every major custom AI ASIC house. The switching cost for a customer to leave Cadence’s toolchain is astronomical—think thousands of engineers retraining, millions of lines of code re-verified, and years of process design kit (PDK) data lost. That stickiness gives Cadence pricing power that rivals even the foundries.
But here is the anomaly: market still values Cadence as a cyclical software vendor. Its revenue growth (15-20% CAGR) and gross margins (88-90%) are superior to most SaaS companies, yet its forward P/E hovers around 30-35x—comparable to a mature enterprise software firm, not an AI infrastructure play. Meanwhile, Nvidia trades at 50x+ despite being a customer that pays Cadence for every GPU design. The CEO’s “undervalued” claim is a call to re-examine the valuation framework.
Core: The On-Chain Evidence—or Rather, the Design Flow Evidence
Let me walk through the data that proves the market is missing the forest for the trees.
First, the AI chip design boom is not a one-time event. Every new generation of AI accelerator requires exponentially more design effort. At 4nm, the total design cost for a complex chip is around $2 billion. At 2nm, that figure rises to $5-7 billion. Of that, 20-30% goes to EDA tools and IP. So for each new GPU or ASIC generation, Cadence collects a larger absolute dollar amount—not just a percentage of rising chip sales, but a rising share of rising design costs. This is a volume-plus-price dynamic that most valuation models miss.
Second, the “AI tax” Cadence collects is not limited to Nvidia. Every hyperscaler—Google, Amazon, Microsoft, Meta—is building custom chips. Every one of those designs passes through Cadence’s flow. The company has zero dependency on any single chip winner. That diversification is a hedge against the inevitable concentration risk in the AI chip market.
Third, the nature of the business is shifting from on-premise license to cloud subscription. Cadence is partnering with AWS, Azure, and Google Cloud to offer EDA-as-a-Service. This expands the addressable market to smaller design houses and startups that could not afford a full license, and it converts revenue from lumpy to recurring. The rule of 40 metrics are already strong: 15-20% revenue growth with 35-40% operating margins. That combination is rare even in software.
Fourth, the R&D intensity is a signal, not a drag. Cadence spends ~30% of revenue on R&D—$12-14 billion annually. That is higher than most software companies. This spending is not just maintaining legacy tools; it is embedding AI into the design flow via the Cadence.AI platform, and acquiring system-level analysis capabilities (e.g., the attempted Ansys acquisition, now pivoted to partnerships). The payoff period is long, but the output will be a larger platform that captures not just chip design but full system design—a TAM expansion from $100 billion to $300 billion.
Contrarian: The Narrative That Correlation Is Causation
Most analysts will tell you that Cadence is a safe bet because AI chip demand is rising. That is true, but it is a lazy argument. The real contrarian angle is that EDA is not a beneficiary of AI chip sales—it is a beneficiary of AI chip design starts, which are less correlated with the chip sales cycle. When AI chip demand slows, design activity actually increases as companies rush to build next-generation alternatives. In the 2022-2023 crypto winter and semiconductor downturn, Cadence’s revenue continued to grow 12-15%. The stock did not correct as much as Nvidia did. That is because EDA is counter-cyclical in a way that market has not fully internalized.
Another blind spot: the bull case for Cadence is often framed as “AI will need more chips, so EDA will grow.” But the opposite is also true: if AI chips become commoditized and margins compress, chip designers will cut costs by reducing design complexity—and that hurts EDA spending. The CEO’s “undervalued” pitch assumes the AI boom will sustain high design complexity, but that is not guaranteed. The real hedge is not the volume of chips, but the stickiness of the toolchain. Even if design starts drop, existing customers are locked in.
Takeaway: The Signal for the Next Quarter
Watch for two things: first, the cloud migration progress. If Cadence’s cloud revenue accelerates beyond 20% of total, it will signal a structural re-rating. Second, listen for any mention of chiplet design tools. The rise of chiplet-based architectures (e.g., AMD’s MI300) requires more IP reuse and verification—a direct boost to Cadence’s IP and system analysis revenue. The market is still pricing Cadence as a software company in a cyclical hardware industry. The data shows it is a platform with a tax on every AI chip ever designed. The code does not lie, only the narrative. And the narrative is due for a correction.