Apple is planning to re-enter the server business in 2029, roughly 18 years after retiring its Xserve line in 2011, with hardware that would run on two or four M8 Ultra chips and may incorporate Nvidia's NVLink Fusion interconnect. The report, from The Information, describes a machine aimed squarely at the AI compute market, where demand for training and inference capacity has outstripped supply across every tier of the stack.
The timing puts the product two full silicon generations away. Apple only announced its M6 chip in August, meaning the M7 has to ship, followed by the M8 family and its high-end Ultra variant, before the server can exist. A 2029 debut is consistent with that cadence but leaves plenty of room for the plan to shift.
The strategic logic is already visible in Apple's current lineup. The Mac Mini and Mac Studio, both built on M-series silicon, have become popular among AI developers who want local inference boxes with strong performance-per-watt, and that popularity has produced shortages. A rack-scale product built on the same ARM-based architecture would let Apple sell into a category where its efficiency story is genuinely differentiated against x86 incumbents.
Key facts
- 01Apple is targeting a 2029 debut for a new server product, its first since retiring Xserve in 2011.
- 02The machines will reportedly run on two or four M8 Ultra chips, depending on the configuration.
- 03Apple only announced the current M6 generation in August, putting the M8 line at least two silicon cycles out.
- 04The servers may incorporate Nvidia's NVLink Fusion, the same tech Nvidia uses to stitch multiple chips into one logical unit.
- 05Apple already turned to Nvidia chips to power its revamped Siri servers, thawing a historically frosty relationship.
The Nvidia angle is the more surprising piece. NVLink Fusion is Nvidia's technology for wiring multiple chips together so they behave as a single accelerator, and licensing it to Apple would mark a notable shift for two companies whose relationship has been frosty for years. That thaw is already underway: Apple recently turned to Nvidia chips to power its revamped Siri servers, ending a long stretch in which the company avoided Nvidia hardware in its own infrastructure.
If NVLink Fusion does end up inside an Apple server, it would let clusters of M8 Ultra chips scale beyond what Apple's own interconnect fabric currently supports. That matters because the M-series has always been optimized for single-socket performance in Macs, not for the multi-chip topologies that AI training runs demand. Borrowing Nvidia's interconnect is a faster route than building one from scratch.
The competitive frame is worth naming plainly. Nvidia dominates AI training silicon, and every hyperscaler is either building custom chips or buying Nvidia by the tens of thousands. Apple entering the server market with ARM silicon puts it in a category currently occupied by Ampere, AWS Graviton, and Nvidia's own Grace CPU, none of which have displaced x86 at scale in enterprise workloads. Apple's advantage, if it has one, is the same integrated hardware-software story that made Apple Silicon successful on the desktop.
There is no indication yet of pricing, form factor, target customer, or whether Apple would sell the machines through its own channel or through partners. The Xserve was a 1U rack unit aimed at small-to-medium businesses and creative shops; a 2029 AI server would need to compete for datacenter budget against very different buyers.
The caveats are real. Rumored 2029 hardware built on chips that have not been announced, using an Nvidia interconnect that has not been publicly licensed to Apple, is closer to a strategic direction than a product roadmap. Plans could change, the M8 Ultra timeline could slip, and the Nvidia partnership could stall before it produces shipping silicon. Apple has also historically preferred to keep its silicon inside its own products rather than sell components or systems into competitive infrastructure markets.
The interesting read is not whether Apple ships an Xserve successor in 2029 but that the company thinks the AI compute shortage is durable enough to justify re-entering a market it walked away from more than a decade ago. That is a bet on AI infrastructure demand remaining structurally undersupplied through the end of the decade, and on Nvidia being willing to license the connective tissue that makes multi-chip Apple servers viable. If both hold, Apple gets a new hardware business. If either wobbles, this becomes another rumored product that never ships.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
