"Most arguments about Apple silicon are two people using one word for three different things."
The Word Everyone Uses
This quest was commissioned by someone who has been against monolithic design for as long as he has been building things, and who still is. That position is the spine of the next track. Before it can be argued, the word has to be taken apart, because "monolithic" describes three different layers of an Apple silicon Mac, and a sentence that is true at one layer is routinely false at another.
Level 1 — The Die
Monolithic die: one piece of silicon holds the whole SoC. Multi-die: several pieces are joined inside one package. Apple's base, Pro and Max chips through the M4 generation are single dies. The Ultra has never been one: since the M1 Ultra (2022), it is two Max dies connected across a silicon interposer Apple calls UltraFusion — "more than 10,000 signals, providing over 2.5TB/s" of die-to-die bandwidth for the M3 Ultra. The M5 generation moved the seam down a tier: M5 Pro and M5 Max are each two dies (the Fusion Architecture), and the M5 Ultra is four. So at the die level, "Apple silicon is monolithic" has been false at the top of the range for four years and is false for the whole pro range now. The industry word for the direction Apple moved in is chiplets; the AMD and Intel versions of the same move appear in the rivals part.
Level 2 — The Package
Monolithic package: the memory is soldered onto the same substrate as the compute, sized once at manufacture. There is no socket, no second slot, no graphics card with its own memory, no external GPU. This is the level where "monolithic" is unambiguously true of every Mac since 2020, and it is the level that carries the cost most people mean: capacity is fixed at purchase, and the only upgrade path is a new machine. It is also the level that carries the benefit: the memory is on a wide, short bus that a socket could not support, which is where the bandwidth in lesson two comes from. The memory track and the homework track both live here.
Level 3 — The Industry
Monolithic industry: one company designs the processor, the operating system, the frameworks and the machine, and sells them as one object. Modular industry: interchangeable parts from many vendors meet at published interfaces — the model the IBM PC established in 1981 and that the Windows-and-Intel stack turned into the world's default. Apple has been the counter-example for forty years, and Apple silicon deepened it: the Mac's processor vendor is now Apple. This is the level where the founder's objection lives, and the level where Christensen's theory of when integration wins and when it loses is the referee.
Modularity Did Not Die; It Moved
Put the three levels side by side and something shows: the industry got more monolithic (level 3) and the package stayed monolithic (level 2) while the die got less monolithic (level 1). The modularity did not disappear. It moved from the user's side of the package — slots, sockets, cards — to the manufacturer's side: dies joined on an interposer, memory stacks chosen per configuration, all of it invisible and unreachable once the lid is on. The M5 Ultra is the most modular Apple chip ever built, at a level no owner can touch. Whether that trade is good is exactly what the next track argues.