macOS Golden Gate, version 27, has quietly shifted how much storage Apple Intelligence and its AI models claim on Macs. Instead of letting users opt out of downloading everything, Apple now mandates model installations during setup—particularly for machines with M3 chips (or newer) and at least 12 GB of unified memory. Models used for Apple Intelligence on those high-end Macs require about 14 GB of on-device storage; other Macs that qualify for Apple Intelligence but don’t meet those specs are officially capped at around 8 GB under Apple’s guidance.
However, actual usage in certain cases dramatically exceeds these numbers. In a Mac mini with an M4 Pro chip, macOS Golden Gate was seen allocating 20.67 GB just for the AI model files. On an M3-based MacBook Air, that figure reached 22.42 GB fresh after installing the beta of Golden Gate. Both of those systems are eligible for Apple’s more powerful “AFM 3 Core Advanced” on-device AI model.
The excess isn’t only on beefy Macs. Older or lower-spec Macs—for which Apple’s rules say just 8 GB is required—are appearing with far more storage consumed. A Reddit user captured a Golden Gate Release Candidate system where Apple Intelligence used 30.16 GB; another reported 24.35 GB used on an M3 Pro MacBook Pro. Even machines running M1 or MacBook Neo chips, technically in the “other” category, have been noted using approximately 14 GB.
What Apple Lists vs What Users See
According to Apple’s documentation, Macs with more modern hardware (M3 or later with ≥12 GB unified RAM) and iPads with the same setup are supposed to allocate up to 14 GB for on-device AI. Other compatible Macs or iPads get up to 8 GB, with Apple Vision Pro and qualifying iPhones following similar rules. So do Apple Watch models, which are limited to about 1.5 GB.
Yet that discrepancy between official “up to” values and real world storage use is wide. The storage readouts in System Settings → General → Storage—under “macOS”—often show numbers far above Apple’s published upper limits. What’s causing this gap is unclear: whether it’s additional cache, different AI model variants, or mis-labelling by the system, doesn’t seem explained in company support documentation.
Importantly, downloading these Apple Intelligence models is no longer optional; toggles to disable them have been removed. Users can’t opt out of setting aside this space on their Macs. You can see how much storage you’re losing by heading to System Settings ➝ General ➝ Storage, then clicking the info icon next to macOS.
Why This Matters
With storage demands from AI growing all the time—with terms like “on-device model,” powerful unified memory, and chip architecture increasingly central—this shift has tangible impact. For users with Macs that have smaller SSDs, what they thought was plenty of free space may now be significantly reduced insightfully. For those who upgraded based on Apple’s published storage requirements, the variance seen in real-world usage could lead to frustration or unexpectedly tight storage margins.
For power users, developers, students, or anyone relying on large file workflows, the cost here is not just gigabytes—it’s potential performance drag, time spent managing space, or even needing to buy external storage. As Apple leans ever harder into on-device AI, expectations for storage capacity and transparency around AI model usage will matter even more.
In short: macOS Golden Gate brings Apple Intelligence fully baked in. Users will likely need to budget well beyond the published 8-14 GB for AI, especially on Macs with higher specs. Going forward, what to watch for is how Apple will explain these gaps, whether future updates will reduce model size or allow more control, and whether these storage requirements will become a standard consideration when buying Mac hardware.
