Lila Soto: Long week — you look like someone who's been reading earnings reports for fun.
Iris Holm: Close. I've been staring at a number and it won't leave me alone.
Iris Holm: Two hundred and seventy-eight million. April 2008. Apple buys P.A. Semi — tiny fabless chip company — and almost nobody covers it as a strategic move. It reads like a talent acquisition. And then everything that comes after it makes you realize that was the actual bet.
Lila Soto: Oh, that's where I want to start — because the P.A. Semi acquisition is kind of the quiet origin of the whole thing we're talking about today, which is Apple's silicon strategy and whether it is actually a moat or just a very expensive head start. Qualcomm and Google's Tensor are closing the benchmark gaps in real-world tests. So is this an architecture story or a timing story?
Iris Holm: That's the frame. And Tim Cook's June 2020 WWDC announcement — two-year transition from Intel x86 to Apple Silicon, done ahead of schedule — that wasn't about raw speed. It was about ownership. Apple now designs the processor, modem, wireless chip, Neural Engine. Intel never got there. Not even close.
Lila Soto: And Intel is the comparison that keeps mattering, I think. Because Intel dominated the PC era and still lost mobile entirely. So what makes this time different — if it is — is really what we're trying to work out.
Iris Holm: Or whether it is different. That's the assumption I want to stress-test.
Lila Soto: And that ownership thing is actually what I want to make concrete — because 'vertical integration' sounds like an MBA term but what it actually means for Apple is weirdly simple. The chip team, the OS team, the services layer — they're not three separate businesses. They're one coordinated stack.
Iris Holm: Here's the analogy. You're building a house. If the architect, the electrician, and the plumber are all on the same team, they route the wiring through the walls before the walls go up. A contractor who hires each one separately — they can't do that. That's Apple versus every Android OEM buying chips from Qualcomm.
Lila Soto: Oh, that's — yeah. Because the OEM doesn't own the wall.
Iris Holm: Exactly. And the silicon version of that is Unified Memory Architecture. CPU and GPU on the same memory pool, same chip package — no duplication, latency basically gone, performance-per-watt jumps. But you can only design that way when your chip team and your OS team are the same people talking every week.
Lila Soto: Wait — who actually runs that? Like who's the person in the room?
Iris Holm: Johny Srouji. Senior VP of Hardware Technologies, teams split between Cupertino and Herzliya. He's the one coordinating across both sites. And the point is — Qualcomm ships a chip. Srouji's team ships a chip that already knows what macOS or iOS is going to ask of it.
Lila Soto: So picture a video editor — say, Saturday afternoon, she's on an M3 MacBook Pro, scrubbing through 4K ProRes footage. The Neural Engine is handling the color grading preview, the GPU's on the timeline render, the CPU's managing the file I/O — and they're all pulling from the same memory pool simultaneously. No bottleneck.
Iris Holm: And a Dell laptop with a Qualcomm chip can't do that — not because the chip is slower, but because the OS and the chip were designed by different organizations. They can't optimize across a boundary they don't both own.
Lila Soto: That's the mechanism. It's not performance specs — it's who designed the wall before the wiring went in.
Iris Holm: And that wall — the chip-OS co-design wall — is only standing because something else is funding it. That's the part I want to pull on.
Iris Holm: Right. Walk me through the actual mechanism.
Lila Soto: Imagine it's Tuesday morning, 8:47 AM. A photographer wraps a late-night edit on her M3 MacBook Pro, AirDrops the RAW file to her iPhone, and her ten years of iCloud photos are already auto-organized, tagged, searchable. Her App Store purchase history — Lightroom, Capture One, whatever — it's just there. She doesn't think 'I'm locked in.' She thinks 'it just works.' That feeling? That's not an accident. That's iMessage, iCloud, Apple Watch pairing, cross-device continuity — all of it quietly raising the cost of ever leaving. And that installed base, over a billion active users, that's what the services layer sits on top of.
Iris Holm: And the revenue number from that layer — fiscal year 2025, Services crossed $100 billion. At gross margins above 73%.
Lila Soto: Which is — I mean, that's the App Store, iCloud, Apple Music, Apple Pay all compounding. And that margin is what funds the $24 billion Apple spent with TSMC in 2025. Up from roughly $2 billion in 2014. Qualcomm can't write that check because Qualcomm doesn't have a services flywheel feeding it.
Iris Holm: And the TSMC relationship has a specific founding moment. Morris Chang committed $10 billion in 20nm capacity to Apple in 2013. That's the line in the sand — that's when Apple bought its way to the front of every future node.
Lila Soto: So the photographer on Tuesday morning — she's not just using a fast chip. She's using the downstream effect of a financial commitment made a decade before she bought it.
Iris Holm: And the part that comes later is where this gets uncomfortable — because that same services layer, the one funding all of it, is exactly what regulators are now circling.
Lila Soto: And that's the crack, right — because Apple doesn't manufacture anything. Every single M-series chip gets made at TSMC in Taiwan. One foundry. One geography. And if that relationship breaks — geopolitical disruption, an invasion scenario, whatever — the $24 billion spend doesn't protect you. It just means you're more exposed.
Iris Holm: That's the inversion. The spend looks like leverage. It's actually dependence.
Lila Soto: And there's no Plan B foundry. Not at that node.
Iris Holm: Now layer the regulatory side on top. The EU Digital Markets Act is targeting the App Store's 30% commission directly — mandatory sideloading, forced lower take rates. That's not hypothetical. That enforcement is active.
Lila Soto: And the U.S. DOJ is running a parallel action. So the same lock-in that generates those 73% services margins — that's exactly what both regulators are calling anticompetitive. I mean, the thing that funds the TSMC relationship is the thing under legal attack.
Iris Holm: Which is — yeah, that's the flywheel breaking. Force open the App Store, compress the take rate below 30%, and suddenly the $100 billion services engine shrinks. Which means the check to TSMC gets harder to write.
Lila Soto: And we've seen this pattern. The 1990s Mac — Apple had genuine architectural superiority then too. End-to-end control. And it lost to the horizontal PC model anyway. So — I guess the question I keep sitting with is, what actually changed between then and now?
Iris Holm: Silicon economics, maybe. Mobile locked people in at the identity layer in a way the PC never did. But — look, I don't think that's settled. The Mac precedent doesn't prove Apple loses. It just proves architectural superiority isn't sufficient.
Lila Soto: So the moat is real — but it has two genuine holes. TSMC is a single geopolitical point of failure, and the services layer funding everything is the exact target regulators chose. Those aren't edge cases.
Iris Holm: Two different things. The architectural advantage is real — it took from P.A. Semi in 2008 to the M1 in 2020, twelve years — and it genuinely exists. But whether it stays profitable is a legal question Apple doesn't fully control.
Lila Soto: Yeah — and I think that's actually the cleanest way I've heard it. The moat is architectural. The revenue is regulatory. Those are two different things sitting on top of each other, and we've been kind of treating them as the same thing this whole conversation.
Iris Holm: And nobody in 2008 wrote the P.A. Semi acquisition as a $278 million bet that would eventually force a $24 billion annual TSMC commitment. It looked like a talent buy. It was the whole architecture.
Lila Soto: And the photographer on Tuesday morning — she still doesn't know any of that. She just thinks it works. Which is maybe the whole story.
Iris Holm: Frankly, that's the most durable part of all of it. Thanks for this.
