And the Winner of Apple's Great AI Bakeoff is... Google

No surprise, but now it's official:
Apple is joining forces with Google to power its artificial intelligence features, including a major Siri upgrade later this year, the tech giants said on Monday.
The multi-year partnership will lean on Google’s Gemini models and cloud technology for future Apple foundational models, according to a statement obtained by CNBC’s Jim Cramer.
Sort of weird that they would announce such a big deal this way rather than official releases/interviews/etc, then again, the talk has been – at least on Apple's side – to downplay the partnership. We get it, it's sort of embarrassing to have to outsource your work in such a key aspect of technology, let alone one you believed you were at the forefront of not that long ago, at least with regard to Siri. And one you promised would get grandma home from the airport soon, only to fail to launch. So now you're stuck outsourcing that work to not just someone else, but one of your chief rivals for years and years. Ouch.
Of course, even as Android battled the iPhone, the two companies remained wedded in Search – one of the most lucrative and divisive deals of all time. And this deal undoubtedly expands upon that one. While they're declining to comment on terms, Mark Gurman of Bloomberg pegged it at around $1B a year back in November. That seems low, especially when we know the Search deal itself is $20B+ a year. But there are a bunch of details we don't yet know. Apple's actual statement on the matter is interesting:
“After careful evaluation, we determined that Google’s technology provides the most capable foundation for Apple Foundation Models and we’re excited about the innovative new experiences it will unlock for our users,” Apple said in a statement.
This makes it sound as if Apple won't run Gemini straight-up, but instead will use Gemini to train (distill?) their own foundational models. Unless those models are really just a white-labeled version of Gemini, which they may be at first.
Another option may be to pipe Gemini into Siri as an option alongside the current ChatGPT partnership – which Apple said isn't changing with this news. Perhaps such integrations are even more pronounced in a future build of iOS this Spring...
If you squint, you can see a three-step strategy from Apple here:
1) Place Gemini as an option for Siri alongside ChatGPT
2) Use different Gemini flavors to help train Apple's own Foundation Models
3) Work on your own custom Foundation Models without Gemini
Again, if they go with #1, I have to imagine they make the placement much more prominent. Even if not in the UI, perhaps they'll just make Siri default to Gemini and/or ChatGPT (depending on which the user chooses?) much more often – perhaps for basically all but the system-level queries (setting timers, etc).
This would buy them some time to work on #2, getting their own new models up to speed. Perhaps for iOS 27 in the Fall, or perhaps even later. Presumably, they'll get first access to new Gemini work from Google to stay at the cutting edge with this deal. And it seems like Apple will probably become one of Google Cloud's biggest partners, if they're not already? As I wrote last November:
There are probably a few other interesting wrinkles in there – one of which may be Apple's willingness to do this because a lot of their cloud infrastructure is already running on Google Cloud. So this may not be as heavy of a lift and as big of an ask as it may seem on the surface. And while the "walled off" aspect is clearly a must for Apple here, you could also imagine that the company may be willing to share some data – fully anonymized, of course – back to help constantly improve the model. And that may speak to why Google would want to do this deal (well, that and the money). Apple has devices in the wild at a scale that basically no one can match. Maybe Samsung, but this potentially unlocks a totally new user base.
Then with #1 and #2, or some combination, #3 would give Apple even more time – years – to completely rebuild and rework their own in-house AI that's less dependent on others. Apple doesn't even have a leader in place for that work yet to replace John Giannandrea and the remaining team has been gutted by Meta so... they need some time. As a bonus, this deal gives access to the best technology right now while they can take their time to figure out if LLMs are fully worth doing on their own, or if other, newer types of models/tech comes into favor...
One more thing: it is interesting that Apple emphasizes the "careful consideration" aspect of the process here. Presumably that means they weighed continuing to do this on their own, but also the possibility of partnering with Anthropic, as has also been reported previously, or going deeper with OpenAI. Not surprising that Apple would go with Google – who yes, just passed them in market cap – over a startup. Still, this must be disappointing to OpenAI given the current partnership. With Anthropic, it just seems like something they only would have done if Apple made it make sense monetarily, which they clearly couldn't agree upon. Back to what I wrote in November:
Apple may not have wanted to pay Anthropic $1.5B a year to use Claude but $1B a year to a partner that is paying you $20B+ for that Search deal? That can be just an in-kind deal! "Google, you know that $25B you owe us this year? Make it $24B, but we'll take a custom build of Gemini. Deal?"
Deal.

Update: There is now an official joint statement, at least on Google's blog.
Similar to the one Apple gave to CNBC, but a bit longer:
Apple and Google have entered into a multi-year collaboration under which the next generation of Apple Foundation Models will be based on Google's Gemini models and cloud technology. These models will help power future Apple Intelligence features, including a more personalized Siri coming this year.
After careful evaluation, Apple determined that Google's Al technology provides the most capable foundation for Apple Foundation Models and is excited about the innovative new experiences it will unlock for Apple users. Apple Intelligence will continue to run on Apple devices and Private Cloud Compute, while maintaining Apple's industry-leading privacy standards.
That last bit seems key. The wording is vague (undoubtedly on purpose), but it seems to suggest that Apple will be able to train Apple Intelligence with Gemini to still run locally on devices – and for the non-local queries, presumably Private Cloud Compute is already running in Google Cloud, even if they don't exactly tout it.









