What this covers
This Business Breakdowns episode walks through Cadence by first explaining the industry it sits under. Jon leads, with Brinton and Matt adding context, and the starting point is EDA: software that chips cannot be designed without, yet a roughly $10 billion market beneath a $550 billion semiconductor industry. The conversation traces how the business works (three-year contracts, per-seat pricing, revenue that follows design starts) and how it was reshaped by Lip-Bu Tan's move to Subscription Revenue Recognition, which also removed the incentive for quarter-end discounting.
The middle of the episode is a primer on chip economics. It separates Fabless designers like NVIDIA from integrated manufacturers like Intel, and explains how rising transistor counts under Moore's Law pushed the industry toward Vertical Disaggregation. It describes designers assembling Off-the-Shelf IP blocks "like Legos," a one-to-two-year redesign cycle tied to new nodes, and the roughly $10 million photomask that makes errors after tape-out so costly. Cadence's emulation hardware, which lets software be tested before silicon exists, is tied to product deadlines such as Apple's September launches.
The later sections look at demand and risk. Systems companies such as Apple, Amazon, and Google now design their own silicon. Brinton recounts Apple's 2008 PA Semiconductor purchase, and Jon cites Amazon's Annapurna Labs team growing from about 100 to over 1,000 people. NVIDIA is also noted as employing more software engineers than hardware engineers. The most contested point is China: Jon argues that homegrown EDA tools are unlikely to catch up quickly given Cadence's 30-plus years of delivered designs. The episode closes on financials, with Cadence reaching the high forties on the Rule of 40, which Jon calls what a mature software business should look like.
Cadence is a structurally advantaged, mission-critical software toll on chip design: its tools are a small share of semiconductor spending, its revenue is recurring and insulated from chip cycles, and its turnaround from a subscription pivot and sustained R&D reinvestment makes it a durable, slow-compounding investment rather than a high-multiple growth bet.
- EDA software is roughly a $10B market sitting beneath a $550B semiconductor industry, so customers treat it as cheap relative to the value it protects
- Subscription contracts, zero churn, and R&D budgets that are cut last make Cadence's revenue cycle-resistant
- Industry consolidation and specialization left a duopoly with pricing discipline, while the top platforms' in-house chip design expands demand
EDA is resilient and durably priced, unlike cyclical chip peers.
- In the semiconductor ecosystem, memory pricing (DRAM and NAND) is the most volatile, followed by capital equipment that can be delayed, and EDA sits at the end of the cut list, so lumping EDA with cyclical chip peers is not a fair comparison.
“So Cadence gets lumped into an offset, EDA gets lumped in with that, sometimes, but it's really not a fair comparison.”
- Cadence creates more value for its customers than it takes, which Brinton calls non-zero-sum, and this makes its pricing durable because customers accept it as a small share of their costs.
“It's really this idea of non-zero sumness, which is a complicated way to say, they create more value than they take”
- Investors can split companies into resilient and optional; optional companies can deliver 10X outcomes, but a resilient company's high-probability 15 percent annual growth, which doubles in five years, is a better bet than a low-probability 10X.
“So, at 2X over five years, with a pretty high probability, is maybe better than a 10X over five years with a pretty low probability.”
- Chips will keep being built regardless of where they are made, because modern electronics, automated safety systems, and digital workflows are not reversible, so the long-term demand trend is clear even with supply-chain and geopolitical disruptions.
“Probably not going to go back to driving cars with carburetors, and no electronics or automatic safety systems. Probably not going to go back to pencil and paper.”
- Chip design is a treadmill that will continue as long as digital progress depends on faster semiconductors, because AI, edge networks, and autonomous driving are all built on the same base layer.
“This treadmill is going to go on for as long as humans continue to progress”
Chip design spreading beyond chipmakers widens EDA's customer base.
- The democratization of chip production enabled large fabless companies like NVIDIA, and it lets firms that never made chips, such as Apple, bring design in-house, a behavior very different from a decade earlier.
“That's also enabling, we talked about Apple at the very beginning becoming a chip maker. That's not possible, also, without this democratization”
- Apple bought PA Semiconductor in 2008 for hundreds of millions of dollars, and that deal grew into Apple's application processors and the M1 chip that displaced Intel in its PCs.
“Apple bought PA Semiconductor in 2008, it was a relatively small transaction, from Apple terms, hundreds of millions of dollars, not billions.”
- Non-technology companies such as Ford are now talking about designing their own chips, a shift Brinton finds surprising, and this widens the customer base for EDA vendors.
“even Ford is talking about designing their own chips, which is like Hell freezing over to me.”
- The goal in licensed IP is to offer off-the-shelf blocks that are easy to resell, since custom IP is owned by the customer and consumes lots of time, so the licensing business has shifted away from custom blocks.
“Really, the goal is to have off-the-shelf IP blocks. In the past, there were a lot of custom IP blocks that were tougher to resell.”
Semiconductor supply layers consolidate into one or two dominant players.
- Across the semiconductor stack, each layer has consolidated to one or two dominant companies, such as one company that makes the latest lithography tools, and EDA has likewise moved from integrated manufacturers into specialized vendors.
“We look down the stack, we keep seeing one or two, maybe three, but usually one or two with the lion's share.”
- Only one company, ASML, makes the extreme ultraviolet lithography equipment used for the latest chips, and in most supply chain layers there are one or two dominant suppliers.
“with what's making the latest chips, it's called EV, there's only one company that does that.”
- Huawei went from being a relatively large chip company to a much smaller one very quickly, illustrating how geopolitical restrictions can rapidly reshape the chip supply base.
“Huawei went from a relatively big chip company, to not a big, big chip company very quickly”
- There is no central repository like GitHub for semiconductor IP; IP is scattered across many companies, so designers must source blocks from several vendors.
“in semiconductors, there is no GitHub of IP. It's distributed around a lot of different companies.”
- The chip ecosystem is becoming more tightly integrated, with chip companies partnering with simulation and verification firms like Ansys and Dassault before tape-out, because mistakes after mask creation are too expensive.
“These chip companies are starting to partner with more simulation companies like Ansys or Dassault... Once you develop that mask... it's too late.”
Long-term customer focus under Lip-Bu Tan contrasts with EDA's weak past.
- Ten years ago EDA businesses were weak, with low operating margins and widespread software piracy, illustrated by a Mentor sales rep who said his software was sold for $15.99 on Chinese shelves.
“There was a lot of piracy with the software.”
- Lip-Bu Tan's team got close to customers, took a long-term view, and invested in R&D roadmaps aligned to customer needs, so that over five years the company can make a large difference even if results are not immediate.
“Lip-Nu came in, and got really close with the customers, really spent a lot of time understanding what the customers needed from Cadence”
- Hardware-in-the-loop simulation, where a chip is emulated on server racks before the physical chip returns from the fab, ensures that when the chip arrives it works with the system integrator's hardware.
“you're doing what's called hardware in the loop simulation, where you're emulating this chip on these racks of servers”