Report: Apple Plans to Make On-Device AI a Key WWDC Focus

[Apple_Robert]

Apple's whole AI thing is so sketchy. First, they massively overpromised, and more than two years later, they still haven't delivered on their promises. Now, it's going to be a watered-down version of what they promised, and their whole Private Cloud Compute seems to have been vaporware all along. Even if they keep the branding, it's not going to happen on their own infrastructure, which is completely different from what they promised. I want less Google deals, not more. On top of that, I am sure there will be a 'Siri+' or something similar to get those sweet, sweet service revenues they're lusting for. I do think the on-device capabilities will be quite limited, yet they're still going to hog all the RAM and battery life, even if you don't want to use them. After all, they needed years just to get simple timers to run on-device with Siri.

The way I see it...we got by fine for decades without AI. Now, so many people feel lost if they can't use AI for tasks they have been doing in their own before AI was even a thing. Keep AI turned off and go about your business on your own and there will be much less friction and disappointment, in my opinion.

 

[svish]

Waiting to see all the AI related updates at WWDC. Apple is currently very far behind and has a lot of catching up to do.

 

[hagar]

Well, there goes my battery life. My iPhone 15 Pro lasts about 3 hours. Lets see how we can get this down even more. With on-device AI.

 

[jchap]

I'm honestly not sure about this. On-device AI capabilities are great and should be the default, but I worry this is going to absolutely wreck battery life.

On-device AI naturally consumes a great deal of power. Sadly, you can't expect long battery life while the GPUs are operating at peak performance. Better to plug in when you need that much power.

 

[eifelbube]

AI tech is rapidly evolving. It's great that Apple is able to quickly pivot and offer AI tech that best meets customer needs while offering the best user privacy. …

“Apple is able to … offer AI tech … ” ?

That Apple “is able to” is only grammatically in the present time — Google, OpenA(s)I(gh), and Anthropic would rightly beg to differ on who is currently delivering the best AI tech. For example, that ImagePlayground has so far been a less capable AI image generation tool than Gemini is proven by the direct comparison in this article. When it comes to AI implementation, a reality check may be necessary for those unconditionally and siriously enthusiastic of everything Apple does.

Maybe Apple will be able to offer AI tech …

 

[Ishimura]

On device AI is the reason (I believe) that I am not able to buy a mini at the moment to replace my aging iMac. Everyone is using them for OpenClaw and other AIs. It's my own fault I waited to update.

No, most people are using those as the local computer harness for very large models that are hosted in the cloud.

 

[Ishimura]

W Apple

Local ai is always the future. Private, cheaper, and doesnt need a harmful datacenter

I’m not sure how you think local (small) LLMs get created without data centers.

 

[Ishimura]

The “efficiency and scale” comes at the expense of communities across the US. Sucking up their water, promising more jobs on paper than needed IRL, and adding noise pollution.

And its also a vehicle for surveillance. You should be in control of your AI, not Scam Altman and his fed cronies.

I know this is an unpopular opinion that few understand, but it is indeed where things are going. Private, decentralized, local AI is the future.

It is part of the future, it is not the future.

There will always be vastly more powerful and capable models hosted in the cloud.

 

[Ishimura]

This is how innovation is supposed to work. If companies all concentrated on making efficiencies in on-device AI instead of building out RAM-sucking data centres that nobody is paying for the market would be in a better place.

They’ll do both. As for the idea that nobody is paying for the product of data centres…lol.

 

[BGrifter]

I think it’s nice Apple waited for the backlash to fully arrive before diving into AI.

 

[Ctrlos]

They’ll do both. As for the idea that nobody is paying for the product of data centres…lol.

They’re not. Subscriptions barely cover the cost of the first few tokens. The vast majority of AI users use the free plans.

Every AI company runs LLMs at a huge loss. OpenAI will end up being bought out by Microsoft. Grok is having to rent out its data centres because not enough people are using them and burning so much capital then its bosses had to merge with SpaceX to siphon funds. Anthropic recently turned a profit it’s true but even that looks like they cooked the books.

Nobody is currently making any money from AI.

 

[I7guy]

They’re not. Subscriptions barely cover the cost of the first few tokens. The vast majority of AI users use the free plans.

Every AI company runs LLMs at a huge loss. OpenAI will end up being bought out by Microsoft. Grok is having to rent out its data centres because not enough people are using them and burning so much capital then its bosses had to merge with SpaceX to siphon funds. Anthropic recently turned a profit it’s true but even that looks like they cooked the books.

Nobody is currently making any money from AI.

I didn’t realize that grok merged with spacex in order to siphon funds from the mothership. That’s interesting news.

 

[Ct applefan]

All I want is Siri to call the right person. I ask Siri call mom. I get “You don’t have a person named bob in your contacts”
I will admit it works well to set a timer or alarm. At least most times.

 

[Happy_John]

Nobody is currently making any money from AI.

Not entirely true. Hardware manufacturers ( nvidia, ram and nand producers etc) are making a hell of a lot of money.

The problem isn’t that “no one” is baking money, it’s that the companies who are making money hand-over-fist are aware but pretending to be unaware that it is simply not sustainable in the current model.

Simply put, companies selling access to their models in data centres are covering costs through investment rounds, as revenue does not cover running costs. Nvidia are selling GPUs and the equipment needed to run hose Goya in data centres, but they’re selling too many chips too quickly, as new data centre builds are going a lot slower than planned - either a lot of the sales are only on paper, or there’s a hell of a lot of gous sitting in storage, already purchased but gathering dust, as there’s not enough data centres to house them.

 

[Ctrlos]

I didn’t realize that grok merged with spacex in order to siphon funds from the mothership. That’s interesting news.

Grok is losing more money than any of the other platforms because it sits idle. X built a massive data centre and ended up having to rent it out to Anthropic so they’re not burning cash just to maintain it.

The average life of the GPUs in these data centres is also 3 years

 

[Ctrlos]

Not entirely true. Hardware manufacturers ( nvidia, ram and nand producers etc) are making a hell of a lot of money.

The problem isn’t that “no one” is baking money, it’s that the companies who are making money hand-over-fist are aware but pretending to be unaware that it is simply not sustainable in the current model.

Simply put, companies selling access to their models in data centres are covering costs through investment rounds, as revenue does not cover running costs. Nvidia are selling GPUs and the equipment needed to run hose Goya in data centres, but they’re selling too many chips too quickly, as new data centre builds are going a lot slower than planned - either a lot of the sales are only on paper, or there’s a hell of a lot of gous sitting in storage, already purchased but gathering dust, as there’s not enough data centres to house them.

At some point those investors are going to want a return and giving away a product is not the way to do it.

Innovation comes not from making your engine bigger but making it more fuel efficient. Anthropic cooked books suggest they’re lowering their compute overheads to make money rather than expanding income.

For the likes of Google and Apple the answer lies in making local models more efficient and reducing the need for the cloud. At some point these models are going to get good enough and data centre usage is going to fall off a cliff.

 

[Happy_John]

At some point those investors are going to want a return and giving away a product is not the way to do it.

Innovation comes not from making your engine bigger but making it more fuel efficient. Anthropic cooked books suggest they’re lowering their compute overheads to make money rather than expanding income.

For the likes of Google and Apple the answer lies in making local models more efficient and reducing the need for the cloud. At some point these models are going to get good enough and data centre usage is going to fall off a cliff.

Yep. It’s not sustainable.

And if OpenAI or Anthropic actually charged a price per token (or million tokens) that did cover running costs, I think the market would balk and wouldn’t.

This is the reason why AI has been pushed as essential over the past years - because the actual cost would be too high to oat unless it was essential, and so you had no choice.

Irrespective of where AI technologies end up going, this “accessing massive, data-centre based models” version of AI is doomed, because of the financial reality: it is very expensive to run.

Personally, I do think were going to end up with far more small, local network models, with specific purposes, but there’s gouging to be a serious financial pain moment between now and then. The amount of capital investment already gone into “big models” means investors will hang on to the “too big to fail” assumption, which will stretch this out and remove any chance of a soft landing.

The “more chips than data centres” is also getting worrying - nvidia have sold a lot of GPUs that may be obsolete by the time there are data centres enough to house them.

 

[name99]

I couldn't care less about AI. I would much rather have Apple put their efforts into fixing bugs and promoting/rescuing Mac gaming (both native development and via emulation like proton/crossover).

Are you seriously not aware that probably the single most economically productive activity of LLMs RIGHT NOW is "fixing bugs"?
What exactly do you think resulted in that massive spike in April?

 

[name99]

Sure they run on phones, but at what speed and quality compared to frontier models running in the cloud? Or something that's as good as, say, Qwen 3.6 27B without needing 128/256GB of memory. Would be happy to hear your thoughts on that.

If only you could easily test this for yourself, at no cost, by following the instructions I gave...

 

[dgdosen]

If only you could easily test this for yourself, at no cost, by following the instructions I gave...

I have! and IMO (everyone has an opinion) for what I want to use it for, local AI is not quite there...

 

[Happy_John]

I have! and IMO (everyone has an opinion) for what I want to use it for, local AI is not quite there...

Not yet. But so much of the focus has been on the “big resource-hungry models”, they’ve effectively sucked a lot of the development oxygen out of the room.

When investors finally draw a line in the sand and stop investing until a clear RoI is visible, that might change, as companies tell development teams to work with what they have, hardware-wiozę.

As the truism goes “Necessity is the mother on invention” - limiting development teams to specific hardware specs, rather than “data centre capacity will increase ad Infinitum” could well result in far mire effective small models.

I personally think the trick with smaller local models will be that they are tailored for specific tasks and workloads, rather than having “one model that does everything”.

 

[seek3r]

apple keeps talking about AI yet they have yet to release anything related to it. apple = vaporware

Apple uses machine learning (aka AI) all over. There’s a lot of ML on Apple’s devices. What you mean is Apple isnt great with LLM chatbots, which… eh

 

[seek3r]

On-device AI naturally consumes a great deal of power. Sadly, you can't expect long battery life while the GPUs are operating at peak performance. Better to plug in when you need that much power.

So do wifi and cellular radios, more cloud processing eats a lot of battery too

 

[huge_apple_fangirl]

Absolutely the right approach by Apple. Right now Apple Silicon Macs + MLX are the best consumer AI platform by far. MLX is getting more and more support in the ecosystem and becoming a first-class target, second to CUDA. But Apple devices are a lot more common than powerful NVIDIA GPUs. Apple is building a platform to run AI on Apple hardware, not a frontier model. The platform is how you make money. Once AI models on the Mac are small and optimized enough, Apple can makes it easy to bring them to the iPhone- via the App Store, and using IAP of course. Each model can be a subscription service, and Apple gets 30%. Apple doesn't need a leading model or agent- there's an app for that. Frontier AI is an app. Apple Silicon + MLX is the AI App Store. Apple is miles ahead of the competition in building out a local AI platform, which is far more important than owning the best model in the leaderboards for a month or two. In a few years, people will be running agents on their MacBook Pros that are good enough to do most of what frontier models today can do, and cost way less.

 

[PBG4 Dude]

Apple uses machine learning (aka AI) all over. There’s a lot of ML on Apple’s devices. What you mean is Apple isnt great with LLM chatbots, which… eh

Still waiting on Windows to be able to copy text from an image. Something iOS/macOS has done for years. That’s like the lowest of low hanging fruit in the ML space.