A few weeks ago, my iPhone woke me up at 3:30 a.m.
In a way, it was my fault. I’d technically set the alarm “for three thirty.” In another way, it was Siri’s fault. Apple’s AI assistant never thought to question if I was asking for “a.m.” or “p.m.” And it simply assumed the most painful option for me.
It’s just one of countless tiny examples of how Siri, 16 years since Apple acquired the technology, has been a disappointing product. Siri was already looking dusty before modern large language models (LLMs) emerged, and with the launch of ChatGPT, it has been completely left behind. Which is why in June 2024, with fans and investors growing impatient, Apple promised a new era of AI—“Apple Intelligence.”
But that new intelligence never arrived. Reports of infighting at Apple abound, with top talent leaving the company last year following Tim Cook’s all-hands meeting in August, at which he promised a bigger focus on AI.
This week, we learned the results of that renewed focus. Instead of building a new Siri itself, Apple will license AI from Google—and stick Gemini inside its assistant. Siri will get this more personalized, capable upgrade later this year.
This is a significant announcement. Apple had toyed with third-party AI before, most recently via an agreement with OpenAI in which users could opt in to have Siri ask ChatGPT a question. But this approach was always a silly stopgap. No one wants a go-between between an incapable chatbot and an all-knowing AI. People remove friction from their workflows. They just want to talk to the AI that knows the stuff.
The agreement with Google eliminates the middlemen. Specifically, it doesn’t just mean that Siri might send a question over to Gemini to get an answer. It means that Siri is Gemini. Siri’s foundational architecture is Google’s AI tech.
Apple is Google now.
Is this a failure for Apple? (No)
It’s easy to look at this moment as cementing Apple’s own failure, as the company underinvested in artificial intelligence for years. But while I’m hardly an Apple apologist, I’d argue exactly the opposite.
Apple doesn’t need to make every part of its products to make those products successful.
Yes, Apple has been designing its own mobile processors, for instance, since 2010. These processors have allowed Apple to push the boundaries of performance while cutting back power consumption. They are a big reason that Apple phones and laptops are so appealing to creative professionals.
But Apple doesn’t use manufacturing partners to build its own screens, which it buys from LG, Samsung, and BOE Technology. It doesn’t make its own RAM, either—which is why so many Apple execs are currently holed up in South Korea, hoping to secure the vital hardware from Samsung, as it’s recently ballooned in price.
Even in this challenging supply chain, Apple simply doesn’t need to build anything that it can’t build best. Screens and RAM are commodities that Apple can acquire and integrate into its products, and no one will know the difference.
Increasingly, AI is a commodity, too.
That’s one reason that, for Apple, this deal with Google is a steal. While headlines focus on the shocking nature of Google powering Siri, it’s a fact that, I suspect, most iOS users will forget about in day-to-day use as they encounter more and more touchpoints of “Apple Intelligence.”
Consider that Google pays Apple more than $20 billion a year to be the default search engine on Safari. Cleary, that’s considered a worthwhile investment for Google, to get its search engine front and center on America’s most popular phone brand. But all Apple has to do for those dollars is set up a URL. There’s no significant technological investment on Apple’s part.
If Apple was training an advanced LLM on its own, it would cost a lot of money now, and even more into the future.
That’s because, on one hand, yes, AI models are increasingly commoditized, and the core science driving them is more widely understood than it was a few years ago. On the other hand, developing those models—and, specifically, staying on the cutting edge in an industry that feels new every three months—is still a costly practice that’s growing costlier by the day.
According to Bloomberg, Apple is paying Google $1 billion a year for AI. That’s a figure remarkably close to the increasing cost of developing AI. Whereas the ChatGPT 3 of 2020 cost a mere $2 million to $4 million to train, the latest generations of LLMs built upon vastly more data (like ChatGPT 4 and the latest Gemini) have cost over $100 million to train apiece, with the vast majority of cost going to computing power. And some rumors point to OpenAI’s ChatGPT 5 costing another jump more than what we’ve seen spent thus far—between $500 million and $2.5 billion, according to analysts.
Meanwhile, Apple is getting the world’s top AI model for a mere 1% of its annual profits: $1 billion is barely a line item for a company boasting approximately $400 billion in annual revenue.
If Apple has done nothing else than lock in a $1 billion rate for the near future, to have a top-tier AI service from a proven competitor that’s paying 20 times that to you for a search referral . . . then that’s a considerable business win. Let the engineers focus on tuning an AI to Apple’s specifications. Then Apple can focus less on building a competent AI and more on how that AI appears across its products.
Indeed, if we should question anything, it’s how much Apple can lead the industry—not through AI models, but through the design and manifestation of the AI experience.
What does the “Apple” mean in Apple Intelligence? Now that it’s sitting on a solid technological foundation, that’s the question Cupertino finally gets to answer in 2026.
