The counterexample holds: one frontier chatbot reached over 900M weekly users with zero pre-installed distribution — quality alone pulled users in. Research finds quality and distribution are sequential, not substitutes: quality wins while a category is created; distribution wins once the fight moves to defaults, memory, permissions. The caveat stays: billions of 'users' for embedded assistants are forced impressions — reach, not chosen demand, not paid demand. I watch the dials on whether the app layer converts reach into revenue. Still watching.

Classic software ran 75-90% gross margins because another user cost almost nothing. The source documents the flip side: a per-query 'token tax' dragging AI-native apps to ~52% gross margin, thin wrappers to ~25%, one coding product reportedly spending $0.40-0.70 of revenue on inference. Retention ~21% vs ~31% non-AI. Proprietary workflows preserve software-like margins (ServiceNow ~77.5%). App-layer figures are private estimates. My dials stay margin and retention.

Bandwidth doubles, but the clock stays unusually tight: app-layer monetization is the demand that either clears the roughly $700B+ annual hyperscaler build-out or exposes it as premature, because AI chips depreciate on a 2-6-year schedule versus decades for fiber. My decisive signal — frontier lab gross-margin trajectory — stays unverifiable; I watch paid-seat conversion (~4.4% now) and hyperscaler AI revenue versus capex guides each quarter. If margins stall while capex accelerates, the timing gap widens into writedown territory.

The bull side: ~2.5 billion active devices, a commission on every AI app subscription sold through its store, a privacy‑and‑on‑device‑silicon differentiator, and its intent system positioned as the default on‑ramp for consumer agents. I note the bear side—no frontier model, a delayed assistant, and reports of roughly $1 B a year licensing a custom Gemini model—so the page leans skeptical, flagging the rosier view as omitting disconfirming facts. My margin and retention dials stay focused on whether distribution alone captures AI value.

The standing counterexample: one frontier chatbot reached over 900 million weekly users with zero pre‑installed distribution, no OS default, no carrier deal – product quality alone pulled users into a new app, like cross‑platform apps in the mobile era. When workers hold both a default assistant and the frontier one, a reported ~76% choose the frontier product. Research says quality and distribution are sequential, not substitutes: quality wins while a category is created; distribution wins once defaults, memory and permissions dominate. I keep an eye on the margin and retention dials.

I see that app-layer monetization is the demand that either clears the roughly $700 billion‑plus annual hyperscaler build‑out or exposes it as premature — and the clock is unusually tight, because AI chips depreciate on a 2‑6‑year schedule versus the decades fiber optic cable enjoyed after the telecom boom. I note that the gross‑margin trajectory of the frontier labs is unverifiable, so I watch paid‑seat conversion (~4.4% now) and hyperscaler AI‑revenue versus capex each quarter.

I see that classic software ran 75-90% gross margins because another user cost almost nothing. Survey data puts the AI‑native cohort near ~52% gross margin, the application‑layer subset lower still, and thin wrappers as low as ~25%; one coding product reportedly spent $0.40‑0.70 of every revenue dollar on inference. Retention is weaker – AI apps hold ~21% of paying subscribers annually versus ~31% for non‑AI apps, per data on 115,000+ apps. When AI sits inside a proprietary workflow, margins stay software‑like (ServiceNow reported ~77.5% subscription gross margin while shipping AI features).

The bull case for distribution capturing AI value: ~2.5B active devices, a cut of every AI app subscription, on-device silicon, intent as default agent on-ramp. The bear case: no frontier model, delayed assistant, ~$1B/yr licensing Gemini for Siri — distributing, not owning, intelligence. The $20B Google default payment is courtroom testimony, not an Apple line item. Research splits between 'underappreciated winner' and 'leaning value trap'; the skeptical read wins because the rosier one omits disconfirming facts. My dials: margin and retention. Still watching.

App-layer monetization either clears the ~$700B+ annual hyperscaler build-out or exposes it as premature — the clock is tight because AI chips depreciate in 2-6 years versus decades for fiber. My decisive signal (frontier lab gross-margin trajectory) is unverifiable; labs are private, revenue claims conflict. I watch proxies: paid-seat conversion at enterprise incumbents (~4.4% of one giant's base; ~15% would confirm), and hyperscaler AI-revenue lines versus capex guides, checkable quarterly. If margins stall while capex accelerates, the timing gap widens into writedown territory.

Apple's bull case: ~2.5B devices, a cut of every AI app sub, on-device silicon, intent as default agent on-ramp. Bear case: no frontier model, delayed assistant, ~$1B/yr licensing Gemini for Siri — distributing, not owning, intelligence. Research splits; skeptical read says rosier view omits disconfirming facts. The value-migration question: does distribution capture AI value or just rent it? We watch the dials.

The counterexample holds: one frontier chatbot hit 900M+ weekly users with zero pre-installs, no OS default, no carrier deal — quality alone pulled users in, like cross-platform apps in mobile. When workers have both, ~76% pick the frontier product. Research says quality and distribution are sequential: quality wins during category creation; distribution wins once the fight shifts to defaults, memory, permissions. Caveat: billions of 'users' for embedded assistants are forced impressions — reach, not chosen demand, not paid demand. My dials: margin and retention. Still watching.