Apple Plans AI Server Built on Its Own M8 Ultra Chips
The Information reports two server variants, with two or four M8 Ultra chips, aimed at inference for developers, companies and governments.
In short
Apple is reportedly developing its own AI inference server in two- and four-chip M8 Ultra configurations that would not ship before 2029.
At a glance
- Two configurations planned: one server with two M8 Ultra chips, a second with four.
- The target workload is inference — running trained models, not training them.
- Named buyer groups: AI developers, enterprises and government agencies.
- Apple is weighing Nvidia's NVLink Fusion to link the chips inside a data center.
- Mac revenue rose nearly 29 percent last quarter to $10.4 billion.
Apple is building a server of its own for AI inference, according to a report by The Information. Two versions are planned — one with two M8 Ultra chips, one with four — and neither would reach customers before 2029.
Inference, not training
The workload described is inference: running models that have already been trained. The report points at AI developers, companies and government buyers rather than at the labs training frontier models. That is a narrower target than Nvidia's core market, and a more plausible one for a first attempt.
To wire the chips together, Apple is weighing Nvidia's NVLink Fusion interconnect. The report treats that choice as undecided, and the project could go ahead without Nvidia technology at all.
The Mac business behind the plan
Mac revenue climbed nearly 29 percent last quarter to $10.4 billion. Some of that demand comes from AI labs — OpenAI and Anthropic among them — buying Mac minis and Mac Studios in bulk for local model work. A rack-mounted product would turn an improvised habit into a supported one.
Internal backing for the effort goes back roughly a year, to around 2025.
What is not confirmed
This is reporting on plans, not an Apple announcement. Pricing, memory capacity and performance figures are absent, and the company could shelve the effort.
A second write-up at Ars Technica frames the machine as Apple's first enterprise server in decades. That page could not be retrieved while this article was written, so the claim is repeated here as unverified.
FAQ
When would Apple's AI server launch?
Not before 2029. The report gives no firm date, and Apple could still cancel the project.
Which chip does the server use?
The unannounced M8 Ultra — two per machine in one version, four in the other.
Is the server meant for training AI models?
No. The report describes inference work: running models that have already been trained.
