Dec 19, 2022 · 41m · another-podcast
Why are chips interesting again?
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Tech analyst Benedict Evans and host Tony Karen Brown speak with venture partner Jay Goldberg to explore the economic, technological, and geopolitical forces reshaping the global semiconductor industry. The discussion breaks down the slowing of Moore's Law, the rise of custom in-house silicon, and the strategic realities of international chip sovereignty.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 23.9% of the talking time here. How this is scored →
The hosts as informed peer 5.7 Guest teaching 5.7 Guest disagreement 1.3 The hosts pushing back 1.6
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 30:06 Rejection of deterministic macroeconomic catch-up
Jay explicitly rejects Benedict's deterministic thesis comparing China's chip ambitions to Japan and South Korea's historical catch-up.
Hardest push from the hosts ▶ 10:04 Host restates scale thesis against margin reduction assumptionBenedict intervenes to correct Jay's framing, clarifying that his scale argument was about total volume supporting fixed costs rather than per-unit margin savings.
Biggest teaching moment ▶ 26:17 TSMC's true moat explainedJay dismantles the simplistic view that chipmaking is just operating bought machinery by detailing TSMC's non-replicable human capital and defect-debugging workflows.
The host holds their own ▶ 39:00 Host's masterclass on x86 vs. ARM disruptionBenedict takes control to deliver an expert structural breakdown of how ARM overturned x86 dominance via power efficiency and classic market disruption.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Why Microchips Have Returned to the Spotlight | 7 | 2 | 1 | 0 | Benedict frames the entire episode by laying out a comprehensive industry shift from Intel dominance to custom silicon, TSMC fab consolidation, and geopolitical sovereignty. | |
| Fabless Ecosystem, Foundries, and Soaring CapEx Costs | 6 | 6 | 1 | 1 | Benedict highlights the extreme consolidation in leading-edge CapEx, while Jay enriches the framework by detailing the fabless vs. foundry divergence and historic design firm shrinkage. | |
| The Rise of In-House Custom Silicon and ASICs | 6 | 6 | 1 | 2 | Benedict cites Apple's 2006/2007 PA Semi acquisition, and Jay expands on the margin dynamics and software integration driving non-chip companies to design ASICs. | |
| Scale Economics, SoC Architecture, and Computing Growth | 6 | 6 | 2 | 3 | Benedict clarifies his scale thesis after Jay assumes he meant direct unit-cost savings, leading to an analytical discussion on SoC integration and transistor evolution. | |
| Semiconductor Market Cyclicality and Pandemic Supply Shocks | 2 | 7 | 2 | 1 | Tony asks about pandemic bottlenecks, allowing Jay to provide an educational breakdown reframing recent supply shortages as standard semiconductor cyclicality. | |
| The Deceleration and Physical Limits of Moore's Law | 5 | 6 | 2 | 2 | Jay explains the slowdown of Moore's Law, while Benedict introduces a photolithography printing press analogy which Jay playfully qualifies before discussing quantum physical limits. | |
| Architectural Specialization and Novel Material Innovations | 6 | 5 | 1 | 1 | Benedict uses an engineering hammer analogy to discuss the pendulum swing between general compute and ASICs, prompting Jay to detail material innovations like CMOS alternatives. | |
| Geopolitical Realities, U.S. Policy, and TSMC's Arizona Fab | 7 | 7 | 1 | 2 | Benedict contextualizes ASML, Taiwan concentration, and Western subsidies, and Jay delivers an authoritative explanation of TSMC's true moat in yield optimization and human capital. | |
| China's Semiconductor Ambitions and Western Export Controls | 7 | 8 | 3 | 5 | Benedict pushes a macroeconomic catch-up analogy based on Japan and South Korea, which Jay counters by introducing the middle-income trap and TSMC's constantly moving frontier. | |
| Global Semiconductor Posturing vs. Leading-Edge Realities | 6 | 6 | 1 | 1 | Tony and Benedict probe international semiconductor announcements, prompting Jay to distinguish between trailing-edge political signaling and genuine leading-edge capacity. | |
| Semiconductor IP Foundations, ARM, and the Rise of RISC-V | 2 | 7 | 0 | 0 | Jay introduces the IP layer, using an architectural analogy to explain ARM's standard IP licensing model and the disruptive potential of RISC-V in IoT and automotive. | |
| Architectural Disruption: The Transition from x86 to ARM | 8 | 2 | 0 | 1 | Benedict delivers a strong analytical synthesis of the historical transition from x86 to ARM, framing it through classic disruption theory and thermal efficiency. |
Statements from this episode (30)
Evans: Major tech firms are designing proprietary chips in-house
“Everyone from Amazon to YouTube are designing their own chips instead of just buying them off the shelf.”
Goldberg: US and European fabless chip firms plummeted from 2,000 to 200
“20 years ago, there used to be 2000 or so fabless design companies in the US and Europe. Today, there are probably 200 plus a handful of startups.”
Goldberg: Building a modern cutting-edge chip fab costs over $10B
“To build a modern cutting edge fab is, is ten billion dollars easily.”
Goldberg: Only 'one and a half' foundries can produce leading-edge chips
“There's really only one and a half that can really produce at the leading edge, right? TSMC in Taiwan is just way ahead of everybody else. Samsung is kind of the half. They're close, but not quite there. And then there's nobody else.”
Goldberg: Intel fell off the Moore's Law curve around 2012–2014
“And Intel dropped the ball. And we could spend a whole hour talking about Intel and their problems, but yeah, they are, they were once the definition of Moore's law. I mean, it's named after Gordon Moore. He used to be CEO of Intel. He was the founder. And the…”
Goldberg: Most chip companies had 20% to 30% gross margins in early 2000s
“When I started covering the industry in the early 2000, gross margins for chip, most chip companies were in the 30 percents, the 20 percents.”
Goldberg: Consolidated chipmakers now enjoy 60% to 70% gross margins
“There's often only one or two vendors for any given part. And you see, they have, you know, gross margins in the sixties and seventies and operating margins in the twenties.”
Goldberg: Tailoring custom ASICs to software stacks yields superior performance gains
“Chips exist to run software, and if you control your software stack, building a chip that's tailor-made, a special purpose chip, we'll call them ASIC, tailor-made for that software stack, you can get performance gains over general purpose compute, like a sort …”
Goldberg: Qualcomm's Snapdragon processors trail Apple silicon by two years
“And so you look at the current Snapdragon applications processes that Qualcomm makes, it's, I don't know, it's two years behind where Apple's A series is or N series is.”
Goldberg: Designing custom chips yields no direct cost savings versus buying
“And if you actually do the math and sort of figure out cost, cost of designing your own chip versus buying from someone else, it turns out that there is no cost savings directly. You only want to design a chip that conveys some form of strategic advantage, rig…”
Goldberg: Only about 100 non-semiconductor companies can design custom chips
“There's only about a hundred companies in the world that really can design their own chips other than chip companies, right? It's mostly the big internet companies and the big industrial companies.”
Goldberg: Nvidia is writing off tons of GPU inventory
“Today, there's too many GPUs, NVIDIA's writing off tons of inventory.”
Goldberg: Moore's Law has slowed from 18 months to five years
“Moore's law is slowing, right? It is not 18 months anymore. It's five years.”
Goldberg: Chip shrinking remains possible for roughly another 10 years
“There's still, you know, a good, I don't know, 10 years of shrinking possible.”
Goldberg: Cost per transistor is increasing in many cases
“The cost per transistor is actually going up in a lot of cases”
Goldberg: Computing is swinging back from general-purpose to specialized chips
“I mean, it's a pendulum, and we've swung very, very far to general purpose compute, and now we're swinging back towards more specific purpose type chips.”
Goldberg: TSMC's Arizona fab will lag leading-edge chips by half a generation
“And now I think that's actually gonna be a real plant in Arizona. And I think it's gonna have decent capacity enough to sort of satisfy the U.S. Military's needs and Apple's needs. And it will be, I don't think it'll be even a generation behind. I think it'll …”
Goldberg: TSMC's Arizona fab will remain dependent on Taiwan's expertise
“It doesn't mean the U.S. Actually can do advanced chips because that plant is going to be entirely dependent on expertise and knowledge base from the mothership in Taiwan.”
Goldberg: China has failed to copy TSMC by poaching personnel
“And China for years has been trying to poach people out of TSMC to come build their foundries. And it just hasn't worked. Like you, it's something about the whole body of knowledge that's just too, you just can't copy and paste.”
Goldberg: China spent $100B to $200B on semiconductor subsidies over ten years
“And they have spent a hundred billion dollars, two hundred billion dollars over the last 10 years subsidizing the growth of their semiconductor industry.”
Goldberg: China has roughly 2,000 fabless chip design companies
“There are probably 2000 legitimate fabless design chip companies in China today.”
Goldberg: Most Chinese fabless chip companies will not survive
“It's going to shrink. Most of those companies won't survive, but enough will that there's something interesting going on there.”
Goldberg: SMIC remains as far behind TSMC as a decade ago
“The most advanced Chinese foundry, SMIC, Shanghai Manufacturing, is still as far behind TSMC as it was 10 years ago, if not further behind.”
Goldberg: China has produced almost nothing in semiconductor manufacturing equipment
“Their attempts to replicate companies like ASML and LAM and build their own equipment for manufacturing semis has also produced almost nothing.”
Goldberg: New regional semiconductor plants build legacy nodes, not leading-edge chips
“If you dig into the details of what's actually getting built, it's not a leading edge plant. It's not, you know, it's not a three nanometer plant like TSMC is building on. It's a 180 nanometer plant, which is what? 15 years old, 20 years old. And you know, we …”
Goldberg: 90% of annual chip volume uses trailing-edge 80nm+ processes
“The number of chips produced every year, or the wafers produced every year, 90% of them are what we would call trailing edge processes, you know, 80 nanometer and up.”
Goldberg: Cruise autonomous vehicles contain $10,000 to $20,000 in chips
“Because if you look at sort of the autonomous cars that are out there today, like I live in San Francisco and there's autonomous cruise cars driving around. They have 10, 20,000 dollars of chips in them, right?”
Goldberg: True vehicle autonomy requires $1,000 to $2,000 leading-edge chip systems
“All of that has to shrink down. And the only way to sort of be able to deliver true autonomy is sort of a 1002 thousand dollar system, which is almost certainly going to have to have advanced edge, advanced process chips in them.”
Goldberg: RISC-V will probably capture automotive and IoT instead of mobile
“And I think that's going to have important implications, probably not for mobile or compute, But for all the other things that are getting chips built into them, cars, especially cars, home devices, internet of things, all of that is probably going to end up i…”