Apple AI reporting is seamlessly migrating from the tired "Apple is behind" narrative to a new mythology in which the company doesn't just make the ecosystem and AI software, but also makes the silicon used on servers providing some of these next-generation services. So, is Apple getting into servers for AI? 

The Wall Street Journal tells us Apple has a top secret project to develop chips for use in servers. "Project ACDC" (which doesn't involve Australia's top rock band as far as we know) is all about researching high-end Apple chips designed specifically to run AI on server farms. A previous rumor claimed mass production of Apple AI server chips will come in 2025, while Apple's chip partner, TSMC, has allegedly booked its entire production of 3nm chips out to Cupertino for use in Macs, iPhones, iPads and more.

Not everyone accepts this claim. Analyst Mark Gurman has said he does not believe Apple will make its own server chips, "because of cost, lack of a differentiator and on-device nature of its LLM." He might well be right, but if we assume Apple is going ahead with this plan, what motivation would there be?

There are some solid reasons why the company might wish to manufacture the silicon used to drive some of its generative AI (genAI) services, including emissions, privacy, and services. Fleshing those claims out:

Emissions

Right now, genAI demands huge amounts of energy. "Generating 1,000 images with a powerful AI model, such as Stable Diffusion XL, is responsible for roughly as much carbon dioxide as driving the equivalent of 4.1 miles in an average gasoline-powered car," according to MIT Technology Review.

If current speculation is correct and Apple intends to deliver genAI tools for image editing and creation in Photos (and in its Pro apps), then that surge in energy use cannot sit well with the company's carbon emissions targets. One way to dramatically reduce energy consumption? Move to a different processor, and with Apple Silicon promising the most computational performance for the least energy it makes sense for the company to put chips based on the power-sipping tech into its server farms. 

It would make a huge reduction in energy demand.

Privacy

The big problem stopping many enterprises from permitting employees to use genAI for work is privacy. Business users do not want their secrets to slip into the public's view, nor do they want confidential documents shared with servers over which they have no control. This has led many big firms to ban employees from using ChatGPT with company data.

Another response is the evolution of private genAI services, such as the one recently introduced in France by Orange Business and LightOn. By hosting its own cloud-based GenAI-as-a-service on its own servers, Apple could place a mantle of security around those services.

But it can make those services even more secure by using its own built-for-privacy chips to process the data. This would likely reflect the company's commitment around privacy and security, and could conceivably see really important data held only on the device, encoded using iCloud authorization, encrypted and made completely secure. 

If that's true, then Apple's genAI offer becomes a series of useful services that any business professional can safely use; this could form part of a strong and viable response to Microsoft Copilot, powered by OpenAI. After all, open does not beat closed when it comes to confidential or regulated data.

Services

As noted recently, Apple management seems very confident that services income will increase in the current quarter. That confidence must be based on something: either existing services revenue is becoming stable and accelerating, or the company plans to accelerate service take up, possibly with new services. 

While extremely speculative, my thinking given the proximity of WWDC is this: What are the chances Apple's iPad launch might also see the company share a little information concerning some of the work it has been doing in AI, promising to introduce at least some genAI services to iCloud subscribers starting after the developer event? Doing so would potentially ignite service adoption across its platforms, while also putting those services through a very public beta testing period, and potentially further challenging perception the company is behind on AI. 

What about cost?

It's also true that at scale the cost of deploying variants of Apple's existing chips in server farms might end up being far lower than the cost of using third-party processors, given the huge spike in demand for those pieces as every company and its brother works to deploy genAI tools.

Apple might have decided that the cost of building and developing its own chips works out to be more practical in the long term than the cost of buying someone else's and found that by doing so it can differentiate any server-based services it provides as offering "Privacy First" - a very Apple message in my humble opinion, and one that should resonate across every business working with data too confidential to share with other services in the space.

Please follow me on Mastodon, or join me in the AppleHolic's bar & grill and Apple Discussions groups on MeWe.