The Last Inch

A common saying is that the second mouse gets the cheese. The first one springs the trap; the second one walks in and eats. It is the patient case for being a fast follower, and Apple has spent two years making it look like gospel.

By any reasonable scoring, Apple appeared to have lost the AI race. It promised a personal, context-aware Siri at WWDC 2024, then quietly shelved the feature in March 2025. Last month it paid $250 million to settle a class action alleging it had advertised an iPhone capability that did not yet exist. There were rumours of it looking to buy a foundation AI model. And on Monday, Tim Cook walked onto the stage at Apple Park to deliver his final keynote as chief executive before handing the company to John Ternus in September and the headline product was a fix for a thing that was supposed to work eighteen months ago. Even now, Apple would not say when it ships. The new Siri arrived in beta, with a developer release later this year and a consumer version that may not land until 2027.

Apple was not the second mouse. It was the eighth.

And as the eighth mouse it did something stranger than turn up late.

It hired the cat to fetch the cheese.

The new Siri does not run on Apple's intelligence. It runs on Google's. The partnership was signalled in January and confirmed on stage on Monday: the rebuilt assistant leans on a custom Gemini model, reported by Bloomberg at around 1.2 trillion parameters, licensed for roughly $1 billion a year. The heaviest reasoning your iPhone now performs is done by the company whose entire business is the thing Apple has spent fifteen years defining itself against.

Craig Federighi stood on that stage and called privacy in AI "non-negotiable." He said it while introducing an assistant whose intelligence belongs to the largest advertising company on earth.

To Apple's credit, the architecture is real, not theatre. Simple requests stay on the device. Mid-tier requests go to Apple's Private Cloud Compute. Only the heaviest work leaves for Google, and Apple's contract reportedly forbids Google from training on your Siri queries. If you stopped reading here, you would conclude that Apple lost the model race and dressed up the defeat with a privacy story.

That is exactly what I want you to think for one more paragraph. Because it is the trap.

The market's verdict on the day was a shrug and a dip as Apple's stock fell nearly 2%, its worst day in three weeks, because the keynote gave no firm date for when any of this actually ships. That verdict is about timing. Mine is about structure. Because while everyone was scoring the delay, Apple quietly settled the only question that matters.

The core question is this: If the most vertically integrated company in the world, the one that designs its own chips, writes its own operating system, builds its own store and controls its supply chain down to the screw will rent its core intelligence from a rival for a billion dollars a year, what does that tell you about the value of the intelligence? It tells you the model is a commodity. A line item. An input you buy, like electricity.

For three years the entire industry has been scoring the wrong race. We have been watching the benchmarks of whose model reasons better, writes better and codes better as though the smartest model would win. Apple just demonstrated, with a billion-dollar cheque, that you do not need to win that race. You can rent the winner. The intelligence was never the prize.

The prize was the last inch.

The last mile is the logistics problem everyone obsesses over. Getting the product from the warehouse to the door.

The last inch is the distance between a consumer's thumb and the surface that answers them.

Consider what Apple actually owns. More than two billion active devices, every one already in someone's hand, already signed in, already trusted. A new "Search or Ask" panel that routes your question to Siri, to Gemini, or to ChatGPT, with Anthropic's Claude tested as an option too. Apple even named those rivals from the stage, which a company that never names competitors does not do by accident. Every assistant you have heard of is now a choice inside a panel Apple owns. They supply the intelligence; Apple supplies the inch.

Then there is the toll. Apple takes 30% of an AI subscription sold through its store in the first year, 15% thereafter, on revenue it does not train, does not host, and does not support. Its services business just posted a record $31 billion in a quarter, up 16% year on year, now larger than Mac, iPad and Wearables put together. Apple has built a tollbooth on a road other people pay to pave.

And there is the chokepoint. With the new developer tools, your app becomes something Siri can call directly, but only if you have wired it to be callable. If the assistant cannot reach into your app, you do not exist in the agentic store. Apple has quietly moved the thing that matters from "is your app in the store" to "can the agent operate your app." That is a far narrower gate, and Apple holds the key.

The model thinks. Apple decides who gets to reach you through it. That is the last inch.

This reorders more than Apple and Google. It reorders the AI labs themselves and it is worth being precise here, because the obvious reading is not quite the right one.

The bet that now looks most exposed is the consumer one. OpenAI built something genuinely extraordinary: a consumer brand with roughly 900 million weekly users, the fastest mainstream adoption of any product in history. But OpenAI does not own a phone. Its distribution into your pocket runs through Apple's surface and it now sits as one option in Apple's panel, beside the assistant that is already the default. You can build the best-loved consumer AI in the world with what was initially a dramatic first-mover advantage and still be a tenant on someone else's inch.

The bet that looks shrewd in hindsight is the enterprise one. Anthropic passed OpenAI in revenue this spring with a $30 billion run-rate climbing to a reported $47 billion by late May, with around 80% of it from enterprise clients. Enterprise distribution does not run through a thumb press, so it cannot be intermediated by whoever owns the glass.

Now the honest counter-case, because the triumphalist version is wrong. OpenAI raised $122 billion at an $852 billion valuation, and its enterprise revenue is now past 40% and climbing toward parity. And I suspect Anthropic went enterprise less out of prophecy than necessity: it could not out-distribute OpenAI in consumer, so it competed where distribution worked differently. Shrewd in hindsight is not the same as outsmarting by design. There is even a wrinkle that cuts against my own argument: the consumer use Apple is about to absorb runs on Google's brain, so it does not come at Anthropic's expense at all.

But the underlying lesson is that the last inch rewards the company whose growth does not depend on a surface it does not own.

The second mouse gets the cheese. But the eighth mouse arrives so late that the trap is long sprung, the danger long past, the cheese sitting in the open. It doesn't need to be first, or fastest, or even clever. It needs to own the one inch that touches the cheese.

Which is the only reason it can do something this reckless: hire the cat. You let the predator into the house only if you are certain you hold the door. Apple paid the company it has spent fifteen years casting as the thing you need protecting from, the largest advertising business on earth, to power the assistant in your pocket, and called privacy non-negotiable while doing it. It can make that bet because the cat does the thinking, but Apple owns the inch. Google fetches the cheese. Apple decides who gets to eat.

Everyone else is still running the race. Apple is standing at the finish line, collecting a toll on a road it did not build, with a cat it does not own.

That is not losing. That is the most Apple thing imaginable.

Brandflow is written by Justin Billingsley, who has spent his career on all three sides of the industry's table: senior client, global agency leader, technology founder. First published 11 June 2026 in the Brandflow newsletter on LinkedIn.