The last time Apple sold a server to the public, the iPhone 4 was new and the company was still deciding whether the future belonged to phones or to something it had not yet named. The Xserve, a rack-mounted machine aimed at businesses, was quietly discontinued in 2011. Fourteen years later, Apple is drawing up plans to walk back into that market, this time with its own processors inside.
According to people familiar with the matter, Apple is advancing plans for an enterprise server built around its in-house silicon, and the machine would carry networking equipment from Nvidia. The product is aimed at AI developers, other enterprises, and government customers, and it would come in two sizes: a smaller version with two of the planned M8 Ultra chips, and a larger one with four.
The detail that matters is not the chips, which Apple has been building for years, but the connective layer between them. Apple is considering Nvidia’s NVLink Fusion interconnect to tie the accelerators together, the same high-speed technology Nvidia uses to turn racks of its own chips into a single coherent computer. It is a concession that the wiring between chips is a problem Apple would rather buy than spend years solving itself.
The timing is distant and the project could still die. Under current expectations, the server would not reach the market until 2029 at the earliest, and the people familiar with the effort cautioned that it could be terminated before then. Apple has killed ambitious hardware quietly before, and a product this far out carries no commitment.
The push has a champion inside the company. Chief Executive John Ternus, who took over Apple’s hardware engineering a year ago, has been a supporter of the server effort since before his appointment, according to the people familiar with the matter. That matters, because a hardware project of this scale at Apple lives or dies on whether a senior executive is willing to defend it.
The business logic is simple arithmetic. Apple now rents most of its AI compute from cloud providers, paying a markup on every query. If it can serve its own products, and sell capacity to others, on hardware it designs, it stops paying that markup and starts earning from it. The catch is that building data-center machines is a different discipline from building phones, and Apple’s track record there is a decade and a half in the past.
The market has already shown Apple where the demand is. The company’s Mac business brought in a record $10.4 billion last quarter, and the customers are not only consumers. OpenAI has bought tens of thousands of Mac minis and Mac Studios to run its workloads, and Anthropic rents Mac minis through Amazon Web Services. Rising memory prices have even pushed up the cost of older iPhones, a reminder that the component markets Apple depends on are tightening.
That demand is what makes the server plan more than nostalgia. Apple’s silicon has proven itself efficient enough that other AI companies are already paying for it in large volume, just in the wrong form factor. A rack server is the obvious next shape, and the M8 Ultra is the chip that would fill it.
The choice of Nvidia for the interconnect is the most telling detail. Apple has spent more than a decade pulling its hardware stack in-house, component by component, and the instinct to own everything runs deep. To consider an Nvidia part inside a server it designed is to admit that the interconnect is one layer where buying beats building, at least for now.
There is a political dimension as well. Apple has positioned itself as a buyer of American silicon, and a server built from Apple chips and Nvidia networking would let the company claim a domestic supply chain while keeping the core proprietary. In Washington, where data-center buildouts are read as national priorities, that combination travels well.
For now, Apple has said nothing. The company rarely comments on unreleased hardware, and a server that may not ship until 2029 is the kind of detail it has spent years keeping private. The market will read the silence the way it reads every Apple hardware report: as a hint that something real is being built, slowly, behind closed doors, with an executive who believes in it and a competitor’s technology holding the pieces together.
Apple’s competitors have already committed to custom server silicon. Amazon has Graviton, Google has its TPU, and Microsoft has been building Maia accelerators. Apple, which let its server ambitions lapse for a decade and a half, is the last of the big platform companies to re-enter, and the M8 Ultra would be its ticket back. The question that has always followed Apple’s AI buildout is whether it would keep renting from others or build its own, and the server plan, if it survives, answers it in the direction of ownership.
The memory market adds urgency to the arithmetic. The rising cost of memory chips has already pushed up the price of older iPhones, and a server line built on four M8 Ultra chips would consume memory in volume, tying Apple’s data-center economics to the same tight component market it already navigates in its consumer products.
