The Toll Road

Amazon announced it would spend $200 billion on capital expenditures in 2026. The week of that announcement, Amazon, Google, and Microsoft collectively erased $900 billion in market capitalization. Apple, which spent $12.7 billion on capex all of last year, rose 7%.

The market did not punish Apple for losing the AI arms race. It rewarded Apple for refusing to run it. The question is whether that is judgment or luck — and the answer has everything to do with a single economic force that Big Tech's cheerleaders have not yet priced in: the commoditization of AI models.

iOS 27 will open Siri to rival AI assistants — not just ChatGPT. What looked like an admission of weakness is a toll road announcement. Every AI company that wants to reach 1.5 billion Apple device users will now pay Apple for the right. Apple did not lose the AI race. It bought the finish line.

---

The Brief

• The spending divergence is the widest in Big Tech history. Amazon: $200B. Alphabet: $175–185B. Microsoft: ~$145B. Meta: $115–135B. Combined: ~$700B. Apple: $14B. Apple's capex is 2% of its four largest peers combined. In Q4 2025, Apple's quarterly capex declined 19% year-over-year — the only Magnificent Seven company to cut, not grow, AI-related spending. ✓ 24/7 Wall St. · Mar 27 / Fortune · Feb 17

• iOS 27 opens Siri to all AI assistants — that is the toll road announcement. Bloomberg reported Apple will end its exclusive ChatGPT partnership and allow rival AI assistants to integrate with Siri in iOS 27. This transforms Apple from a late AI entrant into the distribution gatekeeper for the entire industry. 1.5 billion devices. The most privacy-trusted consumer platform on earth. Every AI company that wants to reach that audience will now negotiate with Apple. ✓ 24/7 Wall St. · Mar 27

• Foundation models are commoditizing — and that makes the distribution layer the scarce asset. GPT-4 cost $0.03 per 1,000 tokens at launch. The same capability now costs a fraction of that. Claude, Gemini, GPT, Llama, Qwen — all available, all improving, all in price competition. When the models commoditize, the question becomes: who owns the pipe to the consumer? Apple owns the most valuable pipe. ✓ 24/7 Wall St. · Mar 27 / CNBC · Feb 27

• Apple's $130 billion cash position is the strategic weapon its rivals have surrendered. While Amazon, Microsoft, Google, and Meta are loading their balance sheets with infrastructure debt, Apple is sitting on $130 billion in cash and returned $104.7 billion to shareholders in fiscal 2025. If the AI infrastructure buildout produces a glut — as the late-1990s fiber-optic boom did — the company with cash and no stranded assets wins. ✓ MacRumors · Dec 30 / 247wallst · Mar 27

---

The Commodity Trap Everyone Else Is Running Into

There is a pattern that repeats in every technology cycle, and it is running right now in AI. The pattern is called commodity convergence: a new, scarce, expensive capability enters the market. Every major company races to build it. Capital floods in. Capacity explodes. The capability stops being scarce. The price collapses. The companies that spent the most on building it discover their infrastructure is now a cost center competing on price with six identical competitors.

This is what happened to railroad companies that overbuilt track in the 1870s. It is what happened to telecom companies that overbuilt fiber in the 1990s. The infrastructure was real and useful. The returns were not, because the scarcity that justified the investment evaporated.

AI foundation models are exhibiting early signs of exactly this pattern. GPT-4 launched at $0.03 per 1,000 input tokens. Within 18 months, comparable capability was available for $0.002. Meta's Llama is open-source and free. Alibaba's Qwen, DeepSeek, Mistral — all competing on price in a market where costs are falling simultaneously. The companies pouring $700 billion into training the next generation of these models are betting that the scarcity persists. The price trajectories suggest it will not.

"These companies are right back to their comfortable lanes and they are really not directly competing with each other in anything like the way that it looked like they might."

— Craig Moffett, analyst · CNBC · February 27, 2026 ✓ CNBC · Feb 27

If Moffett is right — and the price data suggests he is — then the $700 billion capex cycle is not building a moat. It is building a highway that everyone will use for free. And Apple, which spent its capital on M-series chips and $130 billion in cash reserves, will be the one collecting the toll.

---

What Apple Actually Built While Everyone Was Watching Nvidia

The conventional AI narrative in 2024 and 2025 went like this: OpenAI built the models. Nvidia built the chips. Microsoft, Amazon, and Google built the cloud. Meta built the open-source alternative. Apple fell behind. The narrative was not wrong about what happened. It was wrong about what mattered.

While that race was running, Apple was building something different. 1.5 billion active devices. An operating system with deep hardware integration that no competitor can replicate through software alone. A privacy architecture that has become a genuine consumer preference. A payment system trusted by hundreds of millions. An App Store that processes over $100 billion in transactions annually and takes a commission on every one.

This is the distribution layer. Not the model. Not the data center. The relationship between the consumer and the screen — and the trust that relationship represents. Every AI company needs to reach consumers. Apple owns the most trusted path to the most valuable consumers on earth.

The iOS 27 announcement makes this explicit. By opening Siri to rival AI assistants, Apple is not conceding the AI race. It is monetizing the race. OpenAI, Anthropic, Google DeepMind — they all need distribution. Apple has it. The revenue from that distribution will flow to Apple. The infrastructure cost of running those models will be borne by the model companies. Apple collects the margin without building the road.

---

The Fiber-Optic Parallel — and What Happened to the Companies That Built It

Between 1996 and 2001, American telecom companies spent approximately $500 billion laying fiber-optic cable. The logic was sound: the internet was growing, whoever owned the physical infrastructure would own the future. WorldCom went bankrupt. Global Crossing went bankrupt. Lucent Technologies lost 95% of its market value. The fiber was real. The demand was real. The investment thesis was wrong because the supply-demand math was wrong — too much capacity, followed by a price collapse that destroyed returns on $500 billion in spending.

The company that benefited most from all that fiber was not the companies that built it. It was Google — which used the now-cheap bandwidth to build the most valuable internet business in history. Apple, in the AI era, is the application-layer company with the cash and the relationships to benefit from infrastructure built by others.

The AI capex cycle of 2025–2026 is not identical to fiber. The near-term demand for AI compute is real. But the structural dynamic — massive infrastructure spend, falling costs, uncertain near-term returns, advantage accruing to the application layer rather than the infrastructure layer — is the same.

---

What Happens Next

First, if iOS 27 executes as planned: Apple becomes the most important distribution negotiation in the AI industry. Every AI company's consumer product strategy runs through one conversation: what does Apple charge us for Siri integration? The App Store precedent suggests 15–30%. The market will reprice Apple as an AI gatekeeper the day those economics become clear.

Second, if the AI infrastructure spend produces a glut by 2027–2028: Amazon's $200 billion, Google's $180 billion, and Microsoft's $145 billion in annual capex will be depreciating assets in an oversupplied market. Apple's $130 billion cash reserve becomes the most valuable thing in Big Tech — the ability to acquire at the bottom of a capex cycle, the way the Rockefellers bought railroads after the 1870s crash.

Third, the risk Apple cannot control: on-device AI requires chip performance, and Apple's M-series chips are extraordinary — but the iPhone cycle is the mechanism that puts those chips in 1.5 billion hands. If a competitor builds a device that displaces the iPhone as the primary personal computing surface, Apple's distribution layer erodes. Part 2 of this series examines exactly that risk.

---

The Read

The free market's most reliable signal is price. Foundation model prices have fallen roughly 95% in three years. That is not a sign of a thriving, defensible market — it is a sign of commoditization in progress. When a product's price collapses 95% in three years, the companies whose entire strategy depends on charging for that product have a problem. Amazon, Microsoft, Google, and Meta are building massive infrastructure for a product whose price is in freefall.

Apple's strategy is not restraint. It is precision. The company identified the scarce asset in the AI economy — consumer trust and distribution — and has been quietly accumulating it for fifteen years. Every iPhone sale, every App Store transaction, every privacy-first design decision was infrastructure spending of a different kind: not silicon and fiber, but relationship and trust. You cannot build 1.5 billion trusted relationships by spending $200 billion on data centers. You build them by selling people a product they love, that works, that protects their data. Apple spent thirty years doing that. Now the AI companies need what Apple built.

The $700 billion capex cycle is the most expensive proof-of-concept in corporate history. It will produce real AI capabilities that will generate real economic value. But the returns will not accrue proportionally to the companies that built the infrastructure. They will accrue to the companies that own the distribution layer the infrastructure was built to serve.

Apple's $14 billion in capex and $130 billion in cash is not a sign of falling behind. It is a sign of a company that identified the real scarcity early and spent its capital on the right thing. iOS 27 is not an admission of weakness. It is the announcement that the toll road is open for business. ~ Framework

---

Market Truths · 財經真言 · Published Tuesday, Thursday, Saturday · markettruthspod.com