Apple Inc. is weighing a bold step back into hardware it abandoned 15 years ago. The company is developing enterprise servers built around its own forthcoming processors. These machines would target AI inference workloads that companies prefer to run on premises rather than in public clouds.
Plans remain fluid. Yet people familiar with the discussions say the project has already received support from John Ternus, who backed the idea about a year ago while leading hardware engineering and who now serves as chief executive. The potential product would not reach customers before 2029. It could still be canceled entirely or launched without certain features now under consideration.
Two configurations are under study. One would combine two of Apple’s planned M8 Ultra chips. A more powerful version would pack four. The systems would aim at AI developers, large corporations, and governments seeking efficient, secure on-site computation. Bloomberg reported the core details on Sept. 16, citing sources close to the matter.
But here’s the twist. Apple engineers have concluded that linking multiple M8 Ultra chips at rack scale may require help from an old rival. Discussions have focused on Nvidia’s NVLink Fusion technology. This high-speed interconnect combines hardware connectors and specialized memory. It allows custom chips to integrate smoothly into larger data-center setups dominated by Nvidia gear.
Some staffers view NVLink Fusion as the best available option. Apple’s own interconnect methods could prove too slow or costly at the scale required for enterprise servers. The potential partnership signals a pragmatic thaw in relations between the two companies. Apple recently turned to Nvidia chips to power updated Siri infrastructure. Reuters covered the Nvidia angle the same day.
The move would mark Apple’s first dedicated server product for outside customers since it discontinued the Xserve line in 2011. Back then the company shifted focus toward consumer devices and standard Mac hardware. Demand for its silicon has since exploded in unexpected places. AI developers have snapped up Mac minis and Mac Studios in huge volumes. Shortages have followed. OpenAI alone bought tens of thousands of the systems.
Apple already builds servers. Its Houston facility produces advanced machines for Private Cloud Compute, the infrastructure behind Apple Intelligence features that require cloud processing. Those servers emphasize privacy. They run only on Apple silicon in Apple’s own data centers or approved partner sites. Some of those servers sat unused on warehouse shelves earlier this year because Apple Intelligence adoption lagged initial projections. 9to5Mac reported the underutilization in March.
By June Apple had expanded Private Cloud Compute to Google Cloud. The setup uses Nvidia GPUs with confidential computing alongside Intel processors and Google’s Titan security chip. Apple devices still trust only cryptographically approved software. The company maintains full control. Yet the hybrid approach shows willingness to tap external infrastructure when its own capacity falls short.
So the new server project builds on existing know-how. It extends the strategy of using in-house silicon beyond consumer products. And it responds to a market shift. Enterprise workloads are moving from training giant models to running inference on them. Efficiency matters more than raw training power in many cases. Apple’s ARM-based designs deliver strong performance per watt. That advantage could appeal to organizations worried about power consumption and costs in their own data centers.
Memory shortages cloud the picture. The entire industry faces tight supplies of high-bandwidth memory chips as everyone races to build AI capacity. Apple’s server ambitions would compete for those same resources. Ars Technica noted the supply-chain pressure on Sept. 16.
Analysts see logic in the timing. Sovereign AI initiatives and on-premises requirements are gaining traction among governments and regulated industries. A secure, efficient Apple server could fit those needs. It might also let the company capture margin on hardware rather than cede it to cloud providers. Yet 2029 remains distant. Much can change. The project carries codename references in some reports, including earlier mentions of M5 or M7 Ultra variants, though current details center on M8.
John Ternus faces his first major strategic test in the CEO role. He took over from Tim Cook on Sept. 1. Hardware has long been his domain. Success here could define his tenure. Failure, or quiet cancellation, would hardly surprise given the preliminary nature of the plans.
Meanwhile demand for Apple hardware in AI circles keeps growing. Developers praise the unified memory architecture and performance of M-series chips for running large language models locally or in small clusters. Turning that success into rack-scale products represents a logical progression. But it requires solving connectivity challenges at scale.
Nvidia opened NVLink Fusion to third parties in 2025. Qualcomm, Arm, Fujitsu, and Marvell have licensed it. Apple would join a growing list if the deal proceeds. The arrangement could even let existing Nvidia customers incorporate Apple servers into their infrastructure. That possibility expands the addressable market.
Privacy remains a selling point. Apple’s Private Cloud Compute model has been audited by independent experts. It promises that data processed in the cloud receives protections comparable to on-device processing. Any commercial server would likely emphasize similar guarantees to differentiate from commodity offerings.
Power efficiency could prove decisive. Data centers strain electrical grids worldwide. Regulators eye new rules on consumption. Apple’s track record with mobile and desktop chips suggests servers that sip rather than guzzle electricity. Such systems might command premium prices from buyers facing both performance and sustainability mandates.
Of course nothing is certain. The Information first broke the story. Multiple outlets quickly confirmed key elements. No official comment has come from Apple. Spokespeople declined to address the reports. Nvidia also stayed silent on potential collaboration.
Still the signals point in one direction. Apple wants greater control over the infrastructure layer. It has invested heavily in its own silicon for more than a decade. Extending that investment into enterprise servers aligns with the bet that custom hardware delivers competitive advantage. The AI boom simply accelerated the timeline.
Whether the product ships in 2029 or later, the mere consideration reveals shifting priorities. Consumer gadgets built the company. Services and silicon now drive growth. Servers could become another pillar. And if Nvidia helps make the numbers work, old rivalries may give way to calculated cooperation.
The industry will watch closely. Developers already reach for Mac hardware when they can get it. Enterprises may soon have a new option for their AI infrastructure. One designed with the same attention to efficiency and security that defines Apple’s consumer lineup. The question is whether that formula translates at rack scale. Early signs suggest Apple intends to find out.
Apple Eyes Return to Server Market With M8-Powered AI Systems first appeared on Web and IT News.