Apple, not Artificial, Intelligence
Just last month, Apple hosted their yearly WWDC - an event where they showcase all the updates to their platforms. Whilst a lot of it is very interesting, and AI centric, I'm going to mostly focus on Private Cloud Compute.
But first…the first half.
WWDC Regular Programming
The first hour of the keynote provided great updates for the Apple ecosystem.
I'm personally excited about Siri getting a huge kick in the pants, and into this decade, plus a bunch of quality-of-life upgrades across each platform, however, like I live-posted during the keynote, the Calculator app with Math Notes automagically calculates equations and graphs, is super cool.
Though, it's not smooth sailing out of the gate…
Touching on Personal AI Assistants, Apple has a huge opportunity with Siri to recapture a lead on the consumer side. Amazon, who have Alexa in so many homes with low-margin speakers, has objectively been getting worse, and below is a great article with some insight into how Amazon's lead in consumer AI Assistant's, declined.
Quick takeaways;
Almost a year ago, Amazon demoed a significant upgrade for Alexa, however due to organization and technological challenges, it's not ready for prime time and is not as advanced as ChatGPT.
Amazon once celebrated for it's speed of innovation, is getting weighed down in bureaucracy and siloed teams, with unrealistic deadlines and lack a clear vision
Apple, not Artificial, Intelligence
The second hour of the Keynote focused on Apple Intelligence. By leveraging the high performing neural engines and acceleration in their own silicon, Apple have essentially got AI at the Edge, your pocket. The reality hits hard though - AI needs fast memory, and for Apple, it's at least 8GB of it.
Apple has limited their on-device AI to any Mac or iPad with a M-series chip (8GB RAM is the minimum spec), the Vision Pro, and the only iPhone that has 8GB of RAM is the iPhone 15 Pro. Any iPhone from 2022 or earlier, any HomePod, any Apple Watch, and any Apple TV, doesn't have close to the memory needed for Apple's AI.
8GB RAM will be the new standard for all iPhone 16's in September, essentially the same minimum that a current MacBook Pro has!
Whilst on-device processing is fast, that RAM can't load massive multi-billion (or trillion) parameter large language or diffusion models, so as necessary, it bursts out these requests to their own Private Cloud Compute (PCC). For supported requests, this will be instead of an external partner like ChatGPT, where you have far less control of your data.
There's evidence Apple's been working on this since at least early 2023.
Private Cloud Compute (PCC)
Focusing on privacy, Apple's developed PCC using their own M-series silicon, which was quite the announcement, though in hindsight, not that surprising.
Microsoft have gone this path with their Cobalt and Maia chips, Google with their TPU's and now Axion, Meta with MTIA, and Amazon with Graviton, Trainium and Inferentia, however each of those companies has decades of experiences running their own, hyperscale cloud platforms.
Given the commoditized nature of their M2-based chipsets (2 years old, and on a mature 5nm processing node), I would expect Apple to use M2 generation hardware, though, they would have to develop much of the software stack for PPC themselves. Like how accommodations needed to be made for M1 based Mac Mini's to be rented through the hyperscalers, for Apple's sake, I would hope PCC would build upon those efforts, possibly using Mac Pro's or their own rack-mounted enclosure, as it would be a herculean endeavor to essentially build a bespoke, ARM-based AI cloud without many existing tools, from scratch.
Whilst Mac Mini's are limited to 10GbE, the very best NVIDIA chips running AI are connected at 400GbE and soon 800GbE, thus if they use Mac Pro's, they are limited to PCIe Gen4 slots and 16 lanes. That interface would be limited to 100GbE or 200GbE network cards, and would need a refresh to Gen5, to get to 400GbE or PCIe 6 for 800GbE, unless they've literally brought Xserve back to life.
Companies like CoreWeave and Lambda have AMD, Intel and NVIDIA + many others to rely on to help build their clouds - whilst not much is known about what partnerships Apple has leveraged here, it's unlikely they are doing everything themselves.
If you think of your iPhone as the kitchen in your home for AI, where you can do most things, PCC is a restaurant, producing a wider variety of meals at higher scale.
Whilst PCC can cook you multiple meals, the ChatGPT integration will give you access to more restaurants and different cuisines, as Apple has also said they will give the user choice on what service to choose (Claude, Llama, etc.).
Of note, is that Apple is offering PCC and ChatGPT for free, without even an account required for ChatGPT. According to 'people briefed on the matter', Apple isn't paying OpenAI anything - presumably this is an exclusivity deal, expiring when a point release of iOS or MacOS, is released, let's say release 18.2, making available alternate LLM integrations.
One big miss is any benefit for business.
Whilst Apple is primarily a consumer-focused device, all of this intelligence could be put to great use when the inputs are business related, and the outputs are productivity focused. Microsoft has Co-Pilot and is shoehorning AI Silicon into PCs for a nice integrated value proposition - Apple could do this too with Apple Intelligence+.
While there is rumor that there will in fact be a paid service (obviously), nothing for Enterprise is mentioned yet.
Whilst I installed the latest developer builds of MacOS and iOS, so you don't have to [spoiler], even in developer beta 2 of iOS18 and MacOS Sequoia, an updated Siri, and Apple Intelligence with ChatGPT integration is not included, so a follow up will be needed once they too become available, in beta, in the coming months.
If you are interested in AI, and need to update your iPhone this year, I'd highly encourage you to check out the WWDC Keynote and let me know your thoughts!