Just a few weeks in the past, my iPhone woke me up at 3:30am.
In a method, it was my fault. I’d technically set the alarm “for 3 thirty.” In one other method, it was Siri’s fault. Apple’s AI assistant by no means thought to query if I used to be asking for AM or PM. And it merely assumed probably the most painful choice for me.
It’s simply one among numerous tiny examples of how Siri, 16 years since Apple acquired the know-how, has been a disappointing product. Siri was already wanting dusty earlier than fashionable LLMs, and with the launch of ChatGPT, it has been utterly left behind. Which is why in June 2024, with followers and traders rising impatient, Apple promised a brand new period of AI—”Apple Intelligence.”
However that new intelligence by no means arrived. Stories of infighting at Apple abound, talent left the company final 12 months as Tim Cook held an emergency all-hands in August promising a much bigger deal with AI.
This week we discovered the outcomes of that renewed focus. As a substitute of constructing a brand new Siri itself, Apple will license AI from Google, and stick Gemini inside its assistant. Siri will get this extra personalised, succesful improve later this 12 months.
It is a important announcement. Apple had toyed with third party AI before, most lately by way of an settlement with OpenAI the place customers might decide in to have Siri ask ChatGPT a query. However this strategy was all the time a foolish stopgap. Nobody desires a go-between between an incapable chatbot and an all-knowing AI. Individuals take away friction from their workflows. They need to simply speak to the AI that is aware of the stuff.
The settlement with Google eliminates center males. Particularly, it doesn’t simply imply Siri would possibly ship a query over to Gemini to get a solution. It implies that Siri is Gemini. Siri’s foundational structure is Google’s AI tech.
Apple is Google now.
Is that this a failure for Apple? (No)
It’s simple to take a look at this second as cementing Apple’s personal failure, as the corporate underinvested in synthetic intelligence for years. However whereas I’m hardly an Apple apologist, I’d argue precisely the alternative.
Apple doesn’t must make each a part of its merchandise to make these merchandise profitable.
Sure, Apple has been designing its personal cellular processors, as an illustration, since 2010. These processors have allowed Apple to push the boundaries of performance whereas chopping again energy consumption. They’re an enormous purpose that Apple telephones and laptops are so interesting to artistic professionals.
However Apple doesn’t use manufacturing companions construct its personal screens, which it buys from LG, Samsung, and BOE Know-how. It doesn’t make its personal RAM, both, which is why so many Apple execs are currently holed up in South Korea, hoping to safe the important {hardware} from Samsung because it’s lately ballooned in value.
Even on this difficult provide chain, Apple merely doesn’t must construct something that it may’t construct greatest. Screens and RAM are commodities that Apple can purchase, combine into their merchandise, and nobody will know the distinction.
More and more, AI is a commodity, too.
That’s one purpose that, for Apple, this take care of Google is a steal. Whereas headlines deal with the surprising nature of Google powering Siri, it’s a undeniable fact that, I think, most iOS customers will overlook about in day-to-day use as they encounter an increasing number of touchpoints of “Apple Intelligence.”
Take into account that Google pays Apple greater than $20 billion a year to be the default search engine on Safari. Cleary, that’s thought-about a worthwhile funding for Google, to get its search engine entrance and middle on America’s hottest telephone model. However all Apple has to do for these {dollars} is about up a URL. There’s no important technological funding on Apple’s half.
If Apple was coaching a complicated LLM by itself, it will price some huge cash now, and much more into the long run.
That’s as a result of, on one hand, sure, AI fashions are more and more commoditized, and the core science driving them is extra broadly understood than it was a couple of years in the past. On the opposite, creating these fashions—and particularly, staying on the leading edge in an trade that feels new each three months—remains to be a expensive observe that’s rising costlier by the day.
Based on Bloomberg, Apple is paying Google $1 billion a 12 months for AI. That’s a determine remarkably near the rising price of creating AI. Whereas the ChatGPT 3 of 2020 price a mere $2 to 4 million to train, the newest generations of LLMs constructed upon vastly extra information (like ChatGPT 4 and the newest Gemini) have cost over $100 million to coach apiece, with the overwhelming majority of price going to computing energy. And a few rumors level to OpenAI’s ChatGPT 5 costing one other soar greater than what we’ve seen spent to this point—between $500 million and $2.5 billion according to analysts.
In the meantime, Apple is getting the world’s high AI mannequin for a mere 1% of its annual income. $1 billion is barely a line merchandise for a corporation with an organization boasting roughly $400 billion in annual income.
If Apple has executed nothing else than lock in a $1 billion fee for the close to future, to have a high tier AI service from a confirmed competitor that’s paying 20x that to you for a search referral…then that’s a substantial enterprise win. Let the engineers deal with tuning an AI to Apple’s specs. Then Apple can focus much less on constructing a reliable AI and extra on how that AI seems throughout its merchandise.
Certainly, if we should always query something, it’s how a lot Apple can lead the trade, not by means of AI fashions, however by means of the design and manifestation of the AI expertise.
What does the “Apple” imply in Apple Intelligence? Now that it’s sitting on a stable technological basis, that’s the query Cupertino lastly will get to reply in 2026.
