Within the first 90 minutes of WWDC 2026, Apple did something very unlike Apple.

It did not define a new device category on its own, did not reinvent an interaction paradigm on its own, and did not lock the entire supply chain inside its own walls.

Instead, it let one of the two largest companies on earth, fold the model of one of the largest AI labs on earth, into its most strategic product.

Three things, all landing in the same compressed window:

  1. Siri is renamed “Siri AI,” and the underlying model is now Google Gemini, directly replacing the ChatGPT handoff layer that lived there for the last cycle.
  2. The next generation of Apple Foundation Models is co-developed with Google, and the full Apple Intelligence stack migrates to a Gemini line.
  3. Siri no longer “hands the question to ChatGPT.” It now consumes context, screen state, and personal data directly, and tightens agentic behavior into the system level.

If you read this as just “Apple picks Gemini,” you will miss the picture.

Put it back into the longer arc from 2024 to 2026, and it is the marker of a formal power reshuffle between Apple, OpenAI, and Google.

I will unpack it in three parts:

  • Part one: why now — why Apple could not delay any longer.
  • Part two: why Google — why OpenAI is out and Google is in, on both geopolitical and compute grounds.
  • Part three: why OpenAI is not fighting — the real signal is the S-1 filed the same day.

Then I will place the three announcements back into the longer AI majors reshuffle.

Part one: why now, and why Apple could not wait

If you look back at 2024 and 2025, the Apple Intelligence story was a long awkward chapter.

When Apple Intelligence launched at WWDC 2024, the market reaction was positive. Apple, for the first time, put a “on-device model + Private Cloud Compute + ChatGPT handoff” stack on the front page. It seemed like the company could keep its privacy story intact and still belong to the AI era.

Then for the next twelve months, something felt more and more off.

WIRED’s piece the day of WWDC 2026, “Apple’s New Siri AI Is Ready to Get Personal,” captures the gap very precisely:

  • Siri’s “personalization” was stuck for years at changing wallpapers and picking playlists.
  • After handoff to ChatGPT, the experience felt stitched: Siri parts were anemic, ChatGPT parts were sharp, and the two halves did not feel like one product.
  • Apple’s on-device model was lagging OpenAI, Google, and Anthropic badly on context length, reasoning depth, and agent planning.
  • Developer enthusiasm for Apple Intelligence was tepid: the API surface was narrow, latency was inconsistent, and a lot of promised features slipped repeatedly.

TechCrunch, in “WWDC 2026: Everything announced on Siri AI, iOS 27, Apple Intelligence and more,” puts it bluntly:

“Foundation” in Apple Intelligence used to be a brand name. With this generation, it actually has a foundation.

That line is sharp, and it is accurate.

“Apple had to swap the base” is, at its core, an internal admission that Apple’s in-house foundation model route could no longer carry Siri as the next-generation entry point by mid-2026.

And Siri is the only consistent entry point across iPhone, iPad, Mac, Vision Pro, HomePod, and CarPlay. If Siri is weak, Apple’s whole AI-era strategy collapses.

That is why the timing is “now.”

It is not that Apple had an epiphany. It is that Apple could not afford to wait one more cycle.

Part two: why Google, and the real reason OpenAI is out

The second sharper question is: why Google, and not Anthropic, Meta, or Mistral?

The answer sits on two layers: geopolitics and compute.

1. The geopolitics: Apple and Google were never enemies

Many people forget that Google has been Apple’s default search engine for a long time.

Safari’s default search, Quick Look’s in-product completions, and several Apple system features have been served by Google since 2005. The renewal of that deal in 2025 keeps the arrangement in the tens of billions of dollars per year range.

In other words, Apple and Google have long been a deep commercial value chain. To read the new partnership as “two companies suddenly teaming up” is to misread the picture.

The more accurate reading is: that value chain just expanded from “search” to “models.”

2. The compute: Google is the only vendor that satisfies three constraints at once

Model vendor selection is not just about who is smartest. You need to satisfy three things at once:

  • Stable large-scale supply: the inference capacity to serve a Gemini-class model at iPhone global scale.
  • On-device / cloud co-design: the willingness to co-build an on-device small model with a cloud large model, the way Apple wants.
  • Enterprise-grade compliance: the willingness to plug into Private Cloud Compute with isolation, auditability, and no-retention guarantees.

Stack those three constraints together, and there are only two vendors in the world who can do it: Google and OpenAI.

OpenAI had been cooperating with Apple on the Siri ChatGPT handoff through 2024 and 2025, and looked like the natural partner to deepen the relationship. But OpenAI was already doing two things:

  • Scaling ChatGPT itself into a standalone consumer product.
  • Preparing for IPO.

Both of those push OpenAI away from being “the supplier behind Siri.” If OpenAI becomes Siri’s foundation, Apple effectively flattens ChatGPT’s capability into the system layer, and OpenAI’s standalone ChatGPT app becomes a thin skin. That is a bad story for an upcoming S-1.

On top of that, OpenAI itself admitted in late 2024 and early 2025 that high-end compute was tight. Allocating scarce capacity to ChatGPT users and OpenAI’s own agent product lines is a more rational use than serving Siri’s full request volume.

So OpenAI’s exit is not a kicking-out. It is a deliberate stepping-aside.

Google’s entry is not just technical. It is the Google Cloud bundle of compute, compliance, and commercial terms.

3. The depth of the deal: not “calling the Gemini API,” but “co-building Foundation Models”

The most under-appreciated part of the announcement: Apple is not just calling the Gemini API.

TechCrunch writes it explicitly:

Apple said it collaborated with Google and the Gemini family of models to develop the next generation of Apple Foundation Models that power its integrated Apple Intelligence experiences.

The keywords are “collaborated to develop.”

The next-generation Apple Foundation Models are themselves co-developed with Google. That means:

  • Model layer: Apple’s on-device base models are switching from fully in-house designs to Gemini-derived distilled and fine-tuned versions.
  • Cloud layer: Apple’s Private Cloud Compute rides on Google Cloud infrastructure.
  • Capability layer: Screen awareness, personal context, and App control are split-built across the two teams.

This is, in effect, Apple outsourcing the “foundation” of its AI strategy.

The political pressure on this decision is huge. Apple has spent more than a decade telling the story of “vertically integrated” — chip to OS to cloud to service, all in-house.

And on one WWDC 2026 night, Apple officially admitted a thing it would never have admitted before: at the foundation-model layer, Apple is not in-house anymore. At least the bottom of the stack is not in-house.

Part three: why OpenAI is not fighting: the S-1 filed the same day is the real signal

On the same day as the WWDC opening, OpenAI officially filed a draft S-1 with the SEC.

That is not a coincidence.

Pieces from Fortune and StartupHub.ai line up roughly as:

  • Valuation band: $852B – $1T
  • Possible listing window: Q4 2026
  • Lead underwriters: Goldman Sachs + Morgan Stanley

For a company at a $1T valuation, the story of “AI infrastructure vendor” is not sexy enough.

The really sexy story is “the entry-point company of the AI era” — a cross-device, consumer-and-enterprise, basic-utility layer of the AI era.

That is why OpenAI is not going to fight for the “Siri foundation” role.

What OpenAI is going to fight for is:

  • The moat around ChatGPT itself as a standalone product.
  • The application matrix of ChatGPT plus vertical agents (coding, office, education, healthcare).
  • Standardization of the developer ecosystem (GPTs, Apps SDK, the Codex route).
  • End-to-end penetration of the enterprise market.

If you stitch those together, you see it clearly: OpenAI is not at the WWDC table. It is at a different table.

That table is called the IPO.

By filing the S-1 the same day as WWDC, OpenAI is telling the market, with action: we are about to be a public company, we are going to be the entry-point company, and we are not going to be the plumber behind someone else’s Siri.

In a sense, that is the most position-aware move OpenAI has made in this whole AI majors reshuffle.

The three announcements, placed back into the reshuffle

If you line up the WWDC three plus the same-day S-1, the mid-2026 AI majors picture sharpens up.

Player Key move Position in the reshuffle
Apple Bring Gemini in to rebuild Siri + Foundation Models Thicken “device + experience,” give up “model layer”
Google Embed Gemini in the largest consumer OS on earth Move from “AI model company” to “AI device + platform company”
OpenAI File S-1, step back from Siri handoff Move from “AI product company” to “AI public company + entry-point company”
Anthropic, xAI, and others Keep grinding on models + tooling Second tier, waiting for the next reshuffle

The most interesting move is the Apple–Google alliance.

For ten years the dominant narrative has framed Apple and Google as AI-era rivals: in-house models versus Gemini, Harvard Business Review pieces on “the two will keep looking more alike.”

But WWDC 2026 exposes a more honest fact: the model layer of this war is not black and white between giants. When in-house ROI can no longer carry the next-generation interaction, letting the strongest external vendor take over the most strategic entry point may simply be the more rational strategic choice.

Apple gave up the “model layer.”

Google picked up the “entry-point layer.”

OpenAI jumped up a layer to the “public company” tier.

The three are no longer competing on the same axis.

This story, told in 2024, would have been almost unbelievable.

By mid-2026, with Siri on stage being rebuilt on Gemini, and OpenAI filing S-1 the same day, it is no longer future speculation. It is a pattern that has already landed.

What this means for normal users and developers

A few practical notes for the people who actually have to live with these shifts.

For iOS and Mac developers

  • The capability boundary of App Intents expands. Siri AI is now built around screen awareness and cross-app action, which means Shortcuts and App Intents finally get a much larger playing field.
  • Be careful with the old Apple Intelligence API. If your app has been calling Apple Intelligence in the last year, do not bet on it. Wait for the new Foundation Models SDK to drop before deciding whether to migrate.

For Android developers

  • Gemini on-device capability will be pushed into a broader device portfolio. The Android ecosystem was already pushing “on-device Gemini,” and this announcement will accelerate that.
  • Cross-device agent frameworks will fragment further. iOS goes down the Apple–Gemini line, Android goes down the Google line, Windows and Web pick their own. The “mediation layer” between these agents is going to be a new race over the next year.

For normal users

  • Siri really is going to get better. Not “slightly smarter.” It is going to actively read the screen, read context, read your calendar and files. The cost is that your personal context is going to flow more deeply into the Apple–Google collaboration chain.
  • In iOS 27, macOS 27, and homeOS, the “AI settings” page matters more than ever. On day one of the upgrade, go into Settings, and turn off anything you do not want shared with Apple Intelligence or Private Cloud Compute.

One last thought

The most expensive line in WWDC 2026’s keynote was not the reveal of any new hardware. It was the moment Apple finally admitted “we are not building the foundation-model layer in-house.”

The earlier that admission comes, the more rational it looks.

For Google, this is a very respectable “stepping out” — from “Gemini is a model” to “Gemini is the iPhone’s foundation.”

For OpenAI, the same-day S-1 is a very respectable “stepping up” — from “AI product company” to “AI entry-point company.”

The second half of 2026’s AI majors reshuffle is no longer a fight over model size. It is a fight over which layer of the stack each giant is willing to own.

Apple, Google, and OpenAI each switched layers.

The next one to switch will be Anthropic, xAI, or Meta. Watch the next quarterly moves closely.

References

  1. TechCrunch, “WWDC 2026: Everything announced on Siri AI, iOS 27, Apple Intelligence and more”: https://techcrunch.com/2026/06/08/wwdc-2026-everything-announced-on-siri-ai-ios-27-apple-intelligence-and-more/
  2. WIRED, “Apple’s New Siri AI Is Ready to Get Personal”: https://www.wired.com/story/apples-new-siri-ai-is-ready-to-get-personal/
  3. TechRadar, “Apple WWDC 2026 as it happened: Siri AI, iOS 27, macOS Golden Gate, and everything announced at Apple Park”: https://www.techradar.com/computing/wwdc-2026-live
  4. MacRumors, “What to Expect From WWDC 2026: Gemini-Powered Siri, iOS 27, macOS 27 and More”: https://www.macrumors.com/guide/wwdc-2026-what-to-expect/
  5. Fortune, “The big questions OpenAI’s trillion-dollar IPO filing may finally answer”: https://fortune.com/2026/05/20/openai-ipo-filing.html
  6. StartupHub.ai, “OpenAI Submits Draft S-1 Filing”: https://startuphub.ai/ai-news/artificial-intelligence/openai-submits-draft-s-1-filing/
  7. The AI Report, “OpenAI files S-1 for IPO at $852B to $1T valuation”: https://www.theaireport.com/articles/openai-s-1-ipo-852b-1t-valuation
  8. Acquired podcast / Acquired Unplugged keynote with Boris Cherny: https://www.acquired.fm/