Apple’s Quiet Comeback: An In-House AI Server, at the Earliest in 2029

Apple may be returning to the server market after nearly two decades away—this time with an eye on the AI business. According to a report from The Information, the company is developing an enterprise server built on its own M8 Ultra chips, intended for sale to AI developers, businesses, and governments. A launch isn’t expected before 2029, and the project could still be shelved before then. What stands out regardless is who’s behind it: Apple’s current CEO, John Ternus, personally championed the project back when he still led hardware engineering.

Key takeaways

• According to The Information, Apple is developing an enterprise AI server with two or four upcoming M8 Ultra chips, built for AI inference rather than training large models.
• For chip-to-chip connectivity, Apple is considering Nvidia’s NVLink Fusion technology—no decision has been made yet.
• A market launch isn’t expected before 2029; the project has been underway for about a year and could still be changed or canceled entirely.
• Momentum is coming from AI firms like OpenAI and Anthropic, which are already buying Mac Studios and Mac Minis in bulk for their own AI workloads.
• Apple already runs its own server infrastructure for Apple Intelligence via Private Cloud Compute—businesses have reportedly asked for access to it and been turned down.

What’s reportedly being planned

At the center are Apple’s upcoming M8 Ultra chips, not yet released. Two versions are reportedly planned: a smaller one clustering two M8 Ultra processors, and a more powerful one using four. Unlike Nvidia’s or AMD’s training clusters, Apple’s server reportedly targets AI inference first and foremost—running already-trained models and generating responses in production, rather than the heavy lifting of training new models from scratch. For fast communication between chips inside a data center, Apple is reportedly considering Nvidia’s NVLink Fusion technology: a bundle of switches, chiplets, and software originally built to connect Nvidia’s own chips but since opened up to third-party processors. Nothing is settled, though—the report explicitly notes Apple could still change course, rely on its own interconnect, or cancel the whole project before 2029.

Why now

There’s a visible trigger behind Apple taking this seriously again: AI labs like OpenAI and Anthropic are already buying Mac Minis and Mac Studios in bulk to run their own AI workloads on Apple’s energy-efficient chips—a side effect Apple’s product planning likely never anticipated. The trend shows up in the numbers too: Apple’s Mac revenue recently grew nearly 29 percent to $10.4 billion. A server purpose-built for the data center would let Apple serve that demand directly and add enterprise features a regular Mac lacks—relevant given how tight compute capacity around new AI models currently is.

The project is also personally tied to John Ternus, who says he backed it about a year ago while still serving as senior vice president of hardware engineering. On September 1, 2026, Ternus formally took over as Apple’s CEO from Tim Cook, who moved into the role of executive chairman after 15 years at the helm. To observers, the fact that Apple’s new chief executive personally championed an AI server project is a signal that Apple under Ternus may give AI infrastructure more weight going forward—though the company has made no public confirmation of the project.

Between Private Cloud Compute and the open server market

Apple isn’t starting from zero. The company already runs its own server infrastructure, Private Cloud Compute, to handle Apple Intelligence tasks too demanding for an iPhone or Mac alone. So far, though, that infrastructure remains purely internal—not offered as a service or hardware to outside customers, even though businesses have reportedly asked for access. Apple’s existing interconnect technology for those servers could become too slow and too expensive once far more chips need to be linked together, which would explain the interest in working with Nvidia—and raises the question of how much third-party infrastructure technology Apple is willing to accept for its own server product, a trade-off cloud providers also wrestle with when it comes to data sovereignty in AI infrastructure. If Apple actually ships these servers, it would mark the first time since the Xserve line ended in 2011 that the company has offered its own chips at scale for standard enterprise infrastructure outside its own services.

Context

For a company long seen as a latecomer in AI, a server like this would be an unusual move—not a race for ever-larger training clusters, but a play in a niche that suits Apple’s strengths: energy-efficient chips for running already-trained models in production. That organic demand from OpenAI and Anthropic for Mac hardware may have provided the spark underscores just how far the AI boom now reaches, even into companies that never set out to profit from it. There’s still plenty of time for course correction before 2029, and The Information itself acknowledges the project could ultimately go nowhere. Any business wanting local, energy-efficient AI hardware today will, for now, still have to rely on off-the-shelf Mac Studio clusters or offerings from other vendors.