An AI Model Designed a Chip in 48 Hours. Cadence's Moat Didn't Crack.

In July, a Chinese AI lab designed a chip in 48 hours without spending a cent on Cadence or Synopsys software. Moonshot AI's Kimi K3 — a 2.8-trillion-parameter open-weight model — ran the entire design flow, from RTL to verified tape-out, on free open-source tools alone. Traders concluded the EDA toll road was being bulldozed and knocked Cadence down nearly 9% in a session. But the fine print: that chip was built on a 45-nanometer node, several generations behind the 3nm and 2nm frontier where the duopoly's tools are welded into the process. The scare was a stress test of the moat, and the moat held.

Here's why that moment mattered. Cadence and Synopsys together control roughly three-quarters of the EDA market, and close to 90% once Ansys is included. Siemens, the distant third, generates barely half of Cadence's revenue. The pair has grown revenue for 13 straight years, through every downturn the industry could produce. That is not luck. That is a toll road.

Now look at the quarter Cadence reported July 27. Revenue rose 24% year over year to $1.58 billion, non-GAAP earnings of $2.11 a share beat estimates, and every product group grew double digits — IP up more than 40%, system design and analysis up 37%, core EDA up 18%. Management then raised full-year guidance by the most in company history: roughly $6.3 billion in revenue, up 19%, with non-GAAP EPS guided to $8.10 and operating cash flow approaching $2 billion. CEO Anirudh Devgan compressed the case into four words: our competitive position has never been better.

The backlog is the quiet number underneath all of it. Cadence ended the quarter with a record $8.1 billion in bookings — roughly 1.3 times annual revenue, meaning over a year of work is already sold. Retention runs above 95% across the core tools and above 99% for signoff and analog. The mechanics explain why: chip design is a sequential pipeline — swap one tool and you re-run every downstream step. Foundries co-develop process kits with the duopoly about 24 months before production, and a modern node carries 25,000-plus design rules. Get any of it wrong and a respin costs $50 million to $100 million. Nobody churns that.

The franchises reinforce the point. Palladium owns 55-60% of the emulation market; Virtuoso is the analog standard, full stop. When a customer's flows, scripts, and training are all sunk into one stack, rival evaluations mostly become pricing leverage — the incumbent answers with a discount, and the evaluation quietly dies.

The AI story runs in both directions, and both favor Cadence. Design for AI: hyperscalers building custom silicon — TPUs, Trainium, Maia, MTIA — added $15-20 billion of fresh design activity in a few years, and systems companies now account for 45% of Cadence's demand. Tooling for a 3nm project costs three to five times what it did at 28nm. AI for Design: the ChipStack AI Super Agent compressed a five-week verification cycle into 24 hours, and the CFO is already structuring three tiers of monetization — subscriptions, usage-based compute, and a virtual engineer agent tier. Selling software that designs chips with fewer humans, at a premium, is the prize.

Cadence is also answering the one threat that matters. Synopsys closed its $35 billion acquisition of Ansys last July; Cadence countered with the €2.7 billion Hexagon D&E deal, completed in February, adding MSC's Nastran and Adams to a multiphysics portfolio aimed at physical AI and automotive. On the foundry front, Cadence now ships certified AI flows for Intel 18A-P and 14A — a historic weak spot — alongside certified TSMC flows from N3 through A14 and a deepened multi-year Samsung 2nm relationship.

The risks deserve names. China is roughly 14% of revenue; export-control penalties topped $140 million in 2025, and Synopsys' China business falling 22% last year is the template for how bad the downside gets. Valuation is not cheap — a forward multiple in the mid-30s with a PEG above 3 — and the K3 episode proved a single demo can erase 9% of market value in a day. The honest long-term question is whether open-source tools and smarter agents eventually route around the toll road at the leading edge. What we know today: the frontier is where the money is, and at the frontier, the toll road is intact.

Put together, Cadence is a business with 82% recurring revenue, gross margins near 86%, roughly $1.6-1.7 billion of free cash flow on about $5.8 billion of trailing revenue, growing 19% inside a two-player industry that AI is making bigger. The Kimi K3 moment was not the beginning of the end. It was a discount on the best-positioned company in the strongest position it has ever described — and the dip buyers have been right.

Disclosure: The Signal holds no position in CDNS. Positions may change. This is not financial advice.