Apple weighs return to server hardware, possibly with Nvidia, to meet AI compute demand

Apple is weighing a return to building its own servers, more than a decade after it killed off the Xserve line, as the company looks to secure the compute it needs for AI workloads. The Information reported the plan, which would put Apple back into hardware it abandoned in 2011, and raised the possibility of a partnership with Nvidia to build the machines. Apple has not confirmed the discussions, and the report did not specify a timeline or which products would run on the hardware.

The company retired the Xserve rack server in 2011 and told customers at the time to migrate to Mac Pro or Mac mini systems, effectively ceding enterprise data-center hardware to Dell, HPE, Supermicro and others. Since then Apple has designed custom silicon for its own consumer devices and leaned on third-party clouds for the heavy lifting behind iCloud, Siri and its other services. Building servers again would mark a sharp reversal of that strategy, driven less by a desire to sell boxes than by the difficulty of buying enough compute.

The AI buildout has strained supply of exactly the components Apple would need. Nvidia's data-center GPUs remain the bottleneck for nearly every large model training run, and lead times for its top-end parts have stretched well into 2025 for many buyers. Pairing with Nvidia could give Apple priority access to accelerators it currently has to compete for, while letting Nvidia tap Apple's manufacturing relationships and design talent. Whether the two companies can agree on terms is an open question, given that they have clashed before over GPU supply and licensing.

Apple has spent years positioning itself as an AI company that keeps processing on-device, a pitch built around privacy. That story gets harder to maintain as models grow and as features like Apple Intelligence lean on private cloud compute, which already routes some requests to Apple-designed servers. If the company controls its own server hardware, it can tune chips and software together the way it does on iPhone and Mac, potentially lowering the cost per query and reducing dependence on outside vendors.

The reported move also lands against a broader shift in which AI demand is pulling companies back into infrastructure they once outsourced. Meta, Microsoft, Google and Amazon have all poured record sums into custom silicon and data centers, and Tesla has built its own training clusters rather than rent them. An Apple server effort, if it materializes, would put the company in that same capital-intensive race, with all the margin pressure that implies for a business used to selling premium devices.

What to watch is whether Apple confirms anything, and whether any partnership with Nvidia extends to chips, full systems or both. The company's next earnings call and its developer conference are the likely venues for any detail, and suppliers in Asia would be the first to signal orders. Until then, the report stands as a signal that even the most consumer-focused giant now sees compute as a problem it may have to solve in-house.