Want to use artificial intelligence (AI) that maintains privacy and doesn’t need a lake full of water or industrial electricity generator to make an emojii? Apple has that.
Stop to consider it: the ongoing AI bubble is totally reliant on investments in server-side infrastructure, water, and energy at colossal levels. While these online services can do what you ask of them, are they really the best articulation of what AI can be?
Dare to be different
I don’t think so, and I believe Apple has a response to that — in part, because of the sheer performance-per-watt offered by Apple Silicon.
Think of it this way: many of the tasks Apple Intelligence can do for you now don’t use any external services at all, don’t require banks of servers, and have no need for additional energy. That’s because they take place on the device, using the energy inside it (and the increasingly sophisticated cooling systems Apple builds).
Those Apple Intelligence features aren’t confined to Apple’s software, either. Thanks to the company’s work on Foundation Models, these large language model (LLM) engines are available for use within third-party apps, too. (Check back next week to learn more about The Omni Group’s work on this and how they’ve found a way to make Apple’s LLMs work in even more personal ways.)
Reach for the stars
Meanwhile, despite well-publicized staff churn within Apple’s AI development teams, the company is working to create other on-device LLM engines for additional tasks. While the tasks supported today may seem a bit limited when compared to the general AI industry hype, Apple has at last lifted off and will introduce additional LLMs over time. These will also be made available to developers and users, putting localized intelligence in every piece of hardware equipped with an Apple processor.
This is all possible because of Apple Silicon, a market-leading processor family that continues to improve regularly and iteratively and is far and away ahead of what others can provide.
This performance is built for AI, and Apple’s making no secret of it. The M5 chip it introduced this week offers four times the performance for AI workloads as the already market leading M4 processor. “M5 ushers in the next big leap in AI performance for Apple Silicon,” said Johny Srouji, Apple’s senior vice president of hardware technologies.
One step beyond
That’s not hyperbole — these improvements make for real results when running AI on Macs (and other Apple devices). It’s already quite possible to run your own local AI on an off-the-shelf Mac. You already do each time you use Apple Intelligence — or when you run LLM models using webAI.
What this means is that while Apple’s home-baked AI might not yet match competitors like OpenAI in terms of what it can achieve, it does have the advantage of not relying on any infrastructure other than the device you already use. You can, of course, access third-party AI services from an Apple device, if you choose.
Don’t stop believin’
So, while the ongoing AI bubble demands billions in infrastructure spending, Apple’s approach empowers the endpoints to run local LLMs as required. That’s important because it also opens up opportunity for agentic, focused AI solutions working together to tackle complex projects. (Does anyone remember SETI at home?)
We are watching the evolution of focused, flexible, AI solutions that run locally on the device, delivering the advantages of Ai without the latency, tokenization or privacy costs. I’m proposing a flotilla of AI-empowered endpoints, each one using as little energy as possible, working together on tasks.
A change is gonna come
That approach has lots of advantages for larger deployments, delivering fixed costs without forcing them to become dependent on cloud-based services. The latter will inevitably need to claw back their vast infrastructure spending with high access charges once the venture capital funding runs out and they need to nickel and dime their users just to survive. Many will not make it.
Think different. Apple’s vision of placing intelligence across all its endpoints will eventually be vindicated by history. And the biggest as-yet unwritten chapter in that story remains to be heard once Siri gains contextual intelligence perhaps next year. If I’m right about this, things are about to get interesting.
You can follow me on social media! Join me on BlueSky, LinkedIn, and Mastodon.
