Apple is back in the server business

Does anyone else out there remember Xserve? 

Discontinued in 2010, this was an Apple server that saw adoption as a supercomputer cluster, and found another use within movie industry workflows as a RAID system. Fans might be interested to know that an Xserve cluster at Virginia Tech ranked No. 7 on the Top 500 list of supercomputers in 2004, topping out at 12.25 teraflops of performance. (That, incidentally, is about the performance of an iPhone 12, or an M1-based Mac.)

Holding it wrong

Apple discontinued the Xserve with a famously terse Steve Jobs email apparently claiming “hardly anyone was buying it.”

Today, with what is arguably the world’s most performant low-power computer chips rolling off production lines, the Apple Silicon opportunity means the company is returning to the server market; it’s tasking Foxconn with making M4-powered servers to run Apple Intelligence as that service gets rolled out globally over the coming year. 

Apple Intelligence servers are currently powered by the M2 Ultra chip, but Apple intends to upgrade these to M4 chips next year. It is alleged that the choice of Taiwan is deliberate, as the company hopes to gain some input from engineers who have worked on Nvidia servers, though as Apple Intelligence is an internal Apple project there’s no conflict of interest in that proposal — at least, not yet.

After all, Apple is not competing in the server market simply by making servers for its own AI, though its M4 Ultra chip might even outperform Nvidia’s mighty RTX 4090 processor, reports claim. So perhaps there’s a pathway there.

Apple now makes servers

Apple uses these servers for Apple Intelligence functions that require more power than the Apple device used to request the task. When those tasks are uploaded to the cloud, they are given to Apple’s own super-private servers or (optionally) outsourced to OpenAI.

To protect the flow of data, the company’s Private Cloud Compute is a server-based Apple Intelligence rig that lets Mac, iPhone, and iPad users exploit Apple’s own AI in the cloud. What’s important about the service is that it maintains the high privacy and security we already expect from Apple. That means Apple won’t get to see or keep your data and will not know what you’ve requested. “Private Cloud Compute allows Apple Intelligence to process complex user requests with groundbreaking privacy,” said Craig Federighi , Apple’s senior vice president of software engineering. 

The idea is that you can use these LLM tools with peace of mind — the kind any rational person will require when handling their own information. I’ve argued before that this is what every cloud-based AI service should strive to deliver, though I don’t think they will; too many business models are based around capturing, exploiting, and even selling information about their users. That’s why some companies ban staff from using AI.

Perhaps it could sell or rent these servers?

The one thing Apple Intelligence has that perhaps isn’t being fully explained is that Apple also offers developers APIs so they can weave the generative AI technology into their products. Right now, that means introducing Apple Intelligence features within them, but given the importance of AI to developers, and the desire among some of them to make smart tools that can be used privately for specific use cases, at what point might Apple offer Private Cloud Compute as a service to provide trusted computing? Perhaps that is why it is putting the system through such rigorous security review?

There has to be an opportunity. There will be some companies who want to make their own AI solutions, but demand the kind of hardcore security Private Cloud Compute provides. Given that Apple has tasked Foxconn with making servers to support that service, at what point will provision of the servers, along with the bare bones, highly secure, software they run, become a business opportunity? There’s a business case, and given Apple is already leading the industry in just how willing it is to open these boxes up for security review, it feels like a potential direction — if there’s any money in it.

And there clearly is — quite a lot, in fact.

As everything becomes AI, where’s the money?

Recognition of the value and need for AI servers is, in part, what has driven Nvidia’s market cap to intermittently overtake that of Apple this year. The need for servers to provide support for AI is a growth opportunity for all in the space — except perhaps for Intel and AMD, who are watching as ARM’s reference designs define expectations for processor performance.

Whether it wants to be or not, Apple is in the server business, and now that it is, it makes sense for the company to generate more revenue from it. After all, who else promises the kind of rock-solid platform-focused security? Who else can provide such fast chips at such low energy requirements? The only snag in this particular ointment is that Apple Intelligence is not inherently cross-platform, though this hasn’t really got in the way of the company’s success for the last couple of decades. 

Please follow me on LinkedIn, Mastodon, or join me in the AppleHolic’s bar & grill group on MeWe.