Apple Spent a Decade Vowing to Build Its Own AI. The New Siri Runs on Google's.

For years the most reliable joke in technology was that you could ask Siri a simple question and get an apology. On Monday at its developers’ conference, Apple finally answered the joke. It unveiled a rebuilt assistant it calls Siri AI — one that can hold a conversation, read what is on your screen, dig an address out of an old text, and sort your photos on command.

The demo was impressive. The two facts surrounding it are more important than anything Apple put on stage, and the company would rather you not dwell on either. The new Siri leans on Google’s artificial intelligence to work at all. And tens of millions of people in Europe and China will not be allowed to use it, by Apple’s own decision, when it ships later this year.

Strip away the keynote polish and those two facts tell a clearer story about where consumer AI is heading than any feature reel: even the most powerful company in the industry could not build a frontier assistant alone, and the fights between platforms and regulators are now settled on the backs of ordinary users who never had a seat at the table.

Apple couldn’t build this one alone

For a decade Apple sold a specific promise about artificial intelligence: it would do the hard work in-house, run as much as possible on the device in your pocket, and treat your data as something to protect rather than mine. That positioning was the whole brand. So the detail buried in Apple’s own announcement is striking. The next generation of Apple Intelligence runs on what the company describes as Apple Foundation Models “custom-built in collaboration with Google and its Gemini models,” with the most demanding requests handled by a cloud model running on Nvidia chips inside Google’s data centers. Bloomberg reported before the event that the arrangement is expected to cost Apple roughly $1 billion a year.

Read that plainly. The company sitting on the largest cash pile in corporate history, that spent years insisting it would rather build than buy, concluded it could not ship a competitive assistant without renting a rival’s brain. Google is not a neutral supplier here. It makes the Pixel phones that compete with the iPhone and the Android software that runs most of the world’s handsets. Apple just made its flagship AI feature depend on that competitor’s models.

This is the part of the AI story that gets lost in the noise about chatbots. The frontier, meaning the largest and most capable models, has consolidated into a handful of companies that now control frontier-scale systems, and everyone else, Apple included, increasingly builds on top of what those few firms produce. The economics point the same direction. The compute bill behind these models is why Nvidia keeps posting record data-center revenue quarter after quarter; the assistant in your next iPhone helps explain the chip company’s valuation. When the most independent player in the business decides dependence is cheaper than going it alone, that is not a footnote. It is the shape of the market.

The regulatory standoff lands on users

The second fact is the one Apple addressed in a separate, defensive statement. When iOS 27 and iPadOS 27 arrive later this year, users in the European Union will not get Siri AI. Not a delay measured in weeks — the marquee features, the dedicated Siri app, the expanded visual tools, the camera integration, simply will not be there for European iPhone and iPad owners.

Apple blames the Digital Markets Act, the EU’s competition law, and says regulators rejected every proposal it offered to ship Siri AI while also opening the door to rival voice assistants the law requires it to accommodate. China is walled off too: Craig Federighi, Apple’s senior vice president of software engineering, said the features will not come there while the company works through regulatory requirements. Europeans will at least get Siri AI on Macs and Vision Pro headsets; on the phone in their pocket, the one device that matters most, they get nothing.

You can assign blame however you like, and the honest answer is that there is plenty to go around. But the people absorbing the cost did not write the DMA and do not run Apple. A European who bought a $1,200 iPhone gets a materially worse product than an American who bought the identical hardware, because a platform and a regulator could not agree on terms. That is a warning about how these fights now resolve. The DMA was written to pry open Apple’s grip on its ecosystem and give consumers more choice. The first visible result is consumers losing access to a flagship feature. Whether that is Apple punishing Europe to make a point or Brussels overreaching, the user is the one left holding a degraded phone.

The counterargument, and why it falls short

Apple has a reasonable-sounding defense for each charge, and it is worth taking seriously. On Google, the company would say it still runs plenty of AI privately and on-device, that sourcing the heaviest models from a partner is ordinary engineering, and that what matters to a customer is whether the assistant works, not whose data center it runs in. On Europe, Apple would say it tried in good faith and that an inflexible regulator forced its hand.

Both arguments have a point, and both dodge the core of the matter. Smart sourcing is still dependence; a feature you cannot deliver without a competitor’s models is a feature you do not fully control, and Apple spent a decade telling customers that control was exactly what set it apart. And “the regulator made us do it” does not change the experience of the European customer one bit. The questions worth asking are not about who is at fault. They are about leverage: who holds it now in the AI supply chain, and who pays when the powers that do hold it cannot agree.

Why this matters beyond one keynote

It is tempting to file all of this under product news and move on. That would miss what the launch actually revealed. The assistant in a billion pockets is becoming a front end for a small number of foundation models owned by a small number of firms, a concentration that raises the same accountability questions now surfacing in court fights over who answers for what AI systems do. And the patchwork of who can use these tools, drawn by regulators and companies rather than by what the hardware can do, is going to keep widening.

Apple fixed Siri. Good. But the price of the fix was a billion-dollar lean on a competitor and a flagship feature withheld from hundreds of millions of paying customers over a regulatory standoff. The demo was the easy part. The dependencies it exposed are the story that will still matter long after the laughter about the old Siri fades.