The Deepdive

Apple's Biggest Admission Yet - Gemini Powers the iPhone

• Allen & Ida • Season 3 • Episode 39

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 12:15

Send a text

A headline that felt impossible just became reality: Apple is partnering with Google to put a custom Gemini model behind the next generation of Siri. We break down the decision with clear eyes—why Apple chose pragmatism over pride, how privacy holds under a shared architecture, and what you’ll actually gain when your assistant stops acting like a command line and starts behaving like a personal AI agent.

We start with the capability gap. Apple’s internal models pushed the limits for on‑device tasks, but they couldn’t deliver the long‑context reasoning and fluid memory that modern workflows demand. Gemini’s custom 1.2 trillion‑parameter model changes the math, enabling richer synthesis across Mail, Messages, Notes, Photos, and the apps you live in every day. Think: pulling your passport number from a photo on request, capturing a new address from a text straight into Contacts, or chaining edits and filing in a single conversation without losing context.

Privacy sits at the center. We walk through Apple’s two‑tiered approach: simple requests handled locally, complex queries routed to Private Cloud Compute, a sealed Apple‑run environment where Gemini executes in a stateless enclave. Your data stays within Apple’s custody, processed transiently and designed for third‑party verification. It’s the same architectural shift now echoing across the industry, as vendors converge on privacy‑first cloud inference to deploy powerful models at scale.

Follow the money and the power. The reported $1B annual AI spend rides alongside Google’s much larger Safari search payments, a case study in co‑opetition under scrutiny. Antitrust remedies force one‑year limits and bar bundling, keeping competition alive and requiring Google to re‑earn placement annually—leaving room for Anthropic or Microsoft if they outpace on quality or cost. We close by asking what this means for Apple’s long‑term roadmap and the rumored Linwood project: is this deep interdependence the new normal, or a smart bridge while the in‑house engine catches up?

If you enjoyed the analysis, follow the show, share with a friend who loves tech strategy, and leave a quick review to help others find us.

Leave your thoughts in the comments and subscribe for more tech updates and reviews.

Allan:

Okay, let's unpack this. On January 12th, 2026, we got a headline that, frankly, felt like a glitch in the simulation.

Ida:

It really did. Apple, the company that basically perfected the walled garden.

Allan:

The absolute champion of it.

Ida:

Yeah, officially partners with Google, its biggest rival.

Allan:

It's just wild.

Ida:

And the deal itself is a monster. Google Gemini is now going to power the next generation of Siri.

Allan:

And this isn't just about Siri getting a little bit smarter. This is a tectonic shift. We're talking about the core intelligence brain of over two billion Apple devices.

Ida:

And that brain is shifting from Apple's own models to a custom-built Google model.

Allan:

Which brings us to the elephant in the room. Why? Why would Apple, of all companies, swallow its pride and basically admit they lost this part of the AI race?

Ida:

It feels like it goes against their entire ethos.

Allan:

It does. So that's our mission for this deep dive. We're going to tackle three big questions. First, why do they do this? Second, how in the world does this work while still, you know, maintaining Apple's whole privacy promise? The big one. And third, what does this tell us about who really holds the power in the AI market right now?

Ida:

The short answer to the why is well, it's pragmatism. They just had to. Apple's own statement talked about a careful evaluation and how Google's tech was the most capable foundation.

Allan:

That's the corporate speak. The reality is they were playing catch-up, and openly so.

Ida:

Right. Remember WWDC 2024, they announced all those amazing personalized Siri features.

Allan:

The ones that needed Siri to actually think.

Ida:

Exactly. And then those features got quietly pushed back to 2026. They basically admitted they couldn't hit the quality standards. Siri was just too mechanical.

Allan:

It couldn't track context. It was a glorified command executor. They needed a quantum leap.

Ida:

And the quantum leap they bought came down to one thing: parameters. The numbers here are um genuinely shocking.

Allan:

Okay, lay them on me.

Ida:

Apple's own internal cloud models were reportedly hovering around maybe 150 billion parameters.

Allan:

Okay.

Ida:

The custom Gemini model they're getting from Google. It is a massive 1.2 trillion parameters. Trillion with the T.

Allan:

That's an eightfold increase. I mean, that's not an upgrade. That's a different species of technology. That's like going from a calculator to a PhD research assistant who's read the entire internet.

Ida:

That is the perfect analogy. The 150 billion model could set a reminder for you. The 1.2 trillion model, that buys you deep memory. It buys you long context reasoning. That's the muscle you need for the heavy lifting.

Allan:

Aaron Powell And heavy lifting in this case means what? Like the AI can follow a conversation.

Ida:

It means the AI can synthesize a whole week of your emails, understand the context from your calendar, and then draft a response in the right tone, all while juggling, you know, five different pieces of information.

Allan:

Aaron Powell That's the capability they bought because they couldn't build it fast enough themselves.

Ida:

Precisely.

Allan:

Okay, so this technical necessity, it must have cost them. The reports I saw said about a billion dollars a year.

Ida:

Around$1 billion annually, yes. Which sounds like a staggering amount of money.

Allan:

It does.

Ida:

But they apparently looked at other partners. Anthropic was on the table, and the rumored price tag for a deal with them was maybe$1.5 billion a year.

Allan:

So Google actually came in cheaper.

Ida:

Not only did they have the better tech, at least in Apple's evaluation, but they also won the contract on price.

Allan:

Okay, but here's the part that is just it's the beautiful absurdity of big tech.

Ida:

Yeah.

Allan:

Apple is paying Google a billion dollars for AI, while Google is already paying Apple somewhere between$18 and$20 billion a year just to be the default search engine in Safari.

Ida:

And that is the absolute key to understanding this entire deal. It's not really a new negotiation. It's just a tiny little extension of an already massive, incredibly lucrative dependency. Google is basically just reallocating a small sliver of the money they already give to Apple.

Allan:

It's the definition of co-opetition, except one side seems to have a lot more leverage.

Ida:

And for you, the person holding the iPhone, this whole thing is designed to be invisible.

Allan:

Completely behind the scenes.

Ida:

The branding is all Apple intelligence. It's Siri. Gemini is just the quiet utility layer they don't talk about.

Allan:

Which of course brings us to the massive multi-trillion dollar question.

Ida:

It's a privacy wall.

Allan:

Yes. How does Apple, the company that built its modern brand on privacy as a competitive advantage, trust Google with our deepest, most personal queries? They can't just be sending my data to Google's cloud. That would be brand suicide.

Ida:

They can't, and they don't. This is where the architecture gets genuinely innovative, uh, if you know, a little complex.

Allan:

Okay, breaking down for me.

Ida:

They're using what they call a two-tiered hybrid model. So tier one, you ask Siri something simple, set a timer for 10 minutes, turn on the flashlight.

Allan:

Basic stuff.

Ida:

Right. That all runs on Apple's smaller on-device model. That request never leaves your phone. It's instant and it's totally private.

Allan:

Okay, that makes perfect sense. Fast and secure for the simple things. But what about the hard stuff? The moment I ask, draft a summary of that long email thread with my bank about my mortgage.

Ida:

Yeah, that's sensitive stuff.

Allan:

That needs the 1.2 trillion parameter brain. That has to go somewhere.

Ida:

Correct. And it gets offloaded. But, and this is the key, it only gets offloaded to Apple's own specialized server infrastructure. They're calling it private cloud compute or PCC.

Allan:

PCC. Okay, so think of it like a secure server farm that Apple built just to run Google's code.

Ida:

That's exactly what it is. These are servers running on Apple Silicon, and they're completely sealed off from the rest of the world.

Allan:

So my sensitive financial data goes through an encrypted channel to an Apple server, Gemini analyzes it, and the answer comes back. But what's to stop a Google engineer from just logging in and watching the data stream? That seems to be the weak point.

Ida:

And that's the guarantee that makes this whole thing work. It's what's called a stateless environment.

Allan:

Stateless.

Ida:

Think of it like this your data is handled on a computer terminal that instantly self-destructs the moment the answer is sent back to you. The data sent to PCC is not stored. It's not made accessible to Google. It is only processed to fulfill that one single request. Then it's gone.

Allan:

That analogy helps a lot. So it's a verifiable, sealed system. The data just evaporates. And I'm guessing it's designed to be auditable so third parties can confirm nothing is being logged.

Ida:

It has to be. That's the only way they can maintain their privacy promise. And what's really fascinating is that this architecture is becoming the industry standard. Oh, really? Yeah. Google launched its own parallel system, private AI Compute, back in November 2025. It uses special hardware enclaves that do pretty much the same thing as Apple's PCC.

Allan:

Seriously. So Google basically copied Apple's playbook for their own devices. That just proves this privacy first cloud model is becoming the only way to handle these powerful AI models at scale.

Ida:

Absolutely. The need for security is now dictating the architecture. So we know why they do it and how they're keeping it private. Let's talk about what this new brain actually does for us.

Allan:

Yes. The functional upgrade.

Ida:

It's huge. Siri is moving from being a rigid instruction executor to a proper personal AI agent.

Allan:

Aaron Powell, I love that phrase personal AI agent because the old Siri was useless if you went off script. What does that unlock in the real world?

Ida:

The aha moments all come from context. Siri can now use knowledge from across your different apps: mail, messages, notes, photos, it can connect the dots. That's what 1.2 trillion parameters gets you.

Allan:

Aaron Powell Give me a good example. Something better than what's the weather.

Ida:

Okay, how about this? You're booking a flight online and the site asks for your passport number. Instead of digging through your files, you just say, need my passport number. Siri can pull that number from a note you saved three years ago, or even from a photo of your passport in your camera roll.

Allan:

Whoa. Now that is genuinely useful. It just connects the data points for you.

Ida:

And it gets even better with what they call on-screen awareness. This is the real agent-like behavior. Okay. Let's say a friend texts you their new address. You don't have to copy it, open contacts, paste it. You just say add this address to their contact cart. Siri understands what's on the screen and it performs the action for you.

Allan:

Aaron Powell So it's not just executing a command, it's making a judgment call based on visible context.

Ida:

A judgment call across different apps on your behalf?

Allan:

Yeah. I think the example that really sold me was the one with photos and notes. You can tell Siri, make this photo pop to enhance it.

Ida:

And then without starting over, you can immediately follow up with now drop it in my summer trip 2026 note.

Allan:

That multi-step cross-app task is the revolution right there.

Ida:

It is. And then you have all the creative tools: image playground for making fun images, image wand for turning a quick sketch into a polished graphic, even genoji for making custom emoji of your family or pets.

Allan:

All coming in iOS 26.4 next spring. It's a massive upgrade. But we can't talk about an Apple Google deal of this magnitude without talking about the regulators.

Ida:

They are standing right over their shoulders, watching every single move.

Allan:

Absolutely. And this whole deal is happening right in the middle of the fallout from the big U.S. antitrust case against Google.

Ida:

You have to remember the context. Judge Meta ruled in August 2024 that Google was an illegal monopolist in search. That ruling forced huge changes in how they're allowed to make these kinds of deals.

Allan:

So the government is basically shaping the terms of this partnership.

Ida:

Directly. The judge's remedies from September 2025 did two crucial things here. First, Google was explicitly barred from compelling partners like Apple to bundle Gemini with their other services.

Allan:

Which means Apple can honestly say they chose Gemini because it was the best tech, not because they were forced to. It preserves the narrative of user choice.

Ida:

Exactly. That was a huge check on Google's power. And the second remedy was arguably even more important. What was that? A one-year limit on any contract for default search or AI placement.

Allan:

So this is not a 10-year lockdown.

Ida:

Not at all. It prevents Google from freezing the market. They have to re-earn that$1 billion contract every single year. It gives competitors like Anthropic or Microsoft a shot to win the deal next year if their tech gets better.

Allan:

So the regulators are actively keeping the market competitive.

Ida:

They're trying to. The market, though, reacted as if this formalizes Google's lead. Alphabet's market cap soared past$4 trillion for the first time since 2019.

Allan:

Of course it did. It's a massive stamp of approval. Gemini's intelligence is now embedded in over 2 billion of the most valuable devices on the planet.

Ida:

It just reinforces the current AI hierarchy. Apple needed the best functionality now. And only Google had it at the right scale and, frankly, at the right price, thanks to that huge Safari deal.

Allan:

So if we zoom out, what does this all mean for the person holding the phone?

Ida:

It means Apple is making a strategic trade. They're prioritizing the superior functionality, the smartness that they get from Gemini.

Allan:

While fiercely protecting their core brand differentiator, the security, by using their own private cloud compute servers.

Ida:

It's the ultimate balancing act, and we really should see this partnership for what it is. It's an interim technology supply chain. It's a bridge.

Allan:

A bridge.

Ida:

Yeah. We know Apple has a long-term project, codenamed Linwood, to eventually build an in-house model that's just as good. They're building their own superhighway, but for now, they had to buy a temporary bridge to keep their users happy.

Allan:

They couldn't afford to be two years behind in the AI arms race. So this collaboration, it brings together the world's two biggest tech platforms, Apple's ecosystem and Google's intelligence.

Ida:

It's simultaneously brilliant and considering their rivalry, completely ridiculous.

Allan:

Which leaves us with a final provocative thought for you to chew on. This huge convergence is happening, but under strict regulatory watch. So is this the new normal, this deep technical interdependence between giants, or is it just a temporary$1 billion a year necessity in a race that's moving too fast?

Ida:

What does it really mean when the best AI brain can no longer be contained within just one company's carefully guarded walled garden?

Head Shot

Allan

Host
Head Shot

Ida

Host