Apple shipped the M1 in late 2020 and stopped selling Intel Macs years ago, so the reasonable assumption is that the move to Apple Silicon finished a while back. The shipped binaries say otherwise.
Of the 10,662 Mac apps we track at the binary level, 16.1% ship Intel-only, with no arm64 slice at all. One in six, more than five years on, running under Rosetta 2 on every Mac Apple sells today.
That is the part of the migration that has not happened, measured from what apps ship rather than what their release notes claim.
We read the architecture from the binary, not the badge
Most "Apple Silicon ready" claims trace back to press releases, store badges, or developer surveys. We read the Mach-O binary inside each shipped bundle and record which CPU architectures it carries: Intel (x86_64), Apple Silicon (arm64), or both in one Universal 2 binary.
A source-repo scanner cannot see this. A repo can target both architectures while the artifact a user downloads carries only one. We look at the slice that ships, which is why our split can differ from a vendor's own description.
Most Mac apps already ship native
- Universal (Intel + ARM): 57.2%. Runs native on both. The compatibility-maximising default, and where most of the apps we track sit.
- Apple-Silicon-only (arm64): 26.7%. No Intel slice at all.
- Intel-only: one in six. No arm64 slice. Rosetta 2 or nothing.
More than eight in ten ship a native Apple Silicon binary of some kind. The transition is the majority position by a wide margin. But "mostly done" is not "done," and the remainder is the interesting part.
The holdouts are the software that has gone quiet
Five years in, an Intel-only slice is less a transitional state than a signal. We cannot read the reason from the binary, so it is worth separating what we measure from what we infer.
What we measure is the slice. What we infer rests on two structural facts. Native toolchains have had first-class arm64 support for years, so the apps least likely to have rebuilt are the older or less maintained ones, where nobody has shipped a fresh binary since the Intel era. And abandonment looks identical to inertia in the binary: an Intel-only slice does not say whether the vendor chose it or has not shipped since 2020. Either way the user is running under Rosetta.
So the long tail of the Mac migration is real and countable. It is one app in six, and the clearest available proxy for software that has stopped moving.
arm64-only is the number to watch
Universal is the conservative default: ship both, break nothing. Dropping the Intel slice is a decision, a developer saying Intel Macs are no longer worth a build. The arm64-only share, currently just over a quarter of the corpus, is therefore the better indicator of how ready developers are to treat Intel Macs as a rounding error.
For anyone shipping developer tools it is also a roadmap input. It tells you how aggressively your customers will drop Intel, and so when you can stop building and testing an x86_64 path without losing them. With well over half still hedging with Universal, the answer today is "not yet." Track the share across snapshots and you have a trigger for that decision rather than a guess.
Minimum-OS targets are a second view of the same question
Architecture is one axis of how modern an app is. The minimum macOS it targets is another. The largest minimum-OS cohorts among the Mac apps we track:
- 10.13 High Sierra: 13.9%, the single largest cohort
- 12.0 Monterey: 12.1%
- 11.0 Big Sur: 10%
- 10.15 Catalina: 9.3%
- 14.0 Sonoma: 5.2%
- 13.0 Ventura: 4.9%
There is a long tail back to 10.11. The largest single cohort still names High Sierra as its floor, a release that predates Apple Silicon. A floor that low can mean an old toolchain or a deliberate choice to keep old hardware in support. The binary does not say which, so it should be checked against the Intel-only set rather than assumed to be the same apps.
Big Sur was the first macOS to run on Apple Silicon, so the Big Sur-and-later cohorts are the Apple Silicon era. But a modern minimum does not force a modern architecture. Monterey and later still run on Intel Macs, so an app can require a recent macOS and still ship Intel-only. Two related views, not one measurement. The runtime and toolchain split of the same corpus is in what 10,778 Mac apps are built with.
How far off "finished" is
A finished transition, read from binaries, would be a corpus with essentially no Intel-only apps: everything Universal or arm64-only, and the arm64-only share climbing as developers retire Intel support.
We are not there. A meaningful share still ships Intel-only, and the Universal majority says most developers are not ready to commit to arm64-only either. We will re-read the binaries at the next snapshot and see how much of that sixth has moved.
About the data
Figures are from Desktop Insights' June 2026 snapshot of more than 25,000 macOS and Windows desktop apps. The architecture split covers the 10,662 Mac apps that carried a detected CPU architecture. Architecture and minimum-OS targets are read from the shipped binary, not from source repositories. Want the architecture dataset for your own market? Ask us.