T1#regulation#privacy

macOS Sequoia and Apple Intelligence — AI Built Into the OS

macOS Sequoia logo
SourceApple Inc. (Wikimedia Commons) · Public domain (below threshold of originality; trademark applies) · View on Commons

Metadata

Date
Decade
2020s
Tier
T1
Sources
14
Connections
05
Tags
#regulation#privacy

On 16 September 2024, Apple released macOS 15 "Sequoia" alongside iOS 18 and iPadOS 18. At the centre of the announcement was a brand for a new bundle of features — Apple Intelligence.

Except that Apple Intelligence was not in the 15.0 that shipped on 16 September. It arrived in stages.

VersionDateWhat landed
macOS 15.0 / iOS 182024-09-16No Apple Intelligence
15.1 / iOS 18.12024-10-28Writing Tools (proofread, rewrite, summarise), notification and mail summaries, redesigned Siri, Clean Up in Photos. U.S. English only
15.2 / iOS 18.22024-12-11Image Playground, Genmoji, Image Wand, ChatGPT integration, visual intelligence
15.4 / iOS 18.42025-03-31Priority notifications, many more languages, rollout to EU iPhone and iPad

The work is split between an on-device foundation model — around three billion parameters, per Apple's own technical report — and Apple-operated Private Cloud Compute servers. On-device inference runs on the Apple Neural Engine, an NPU: a processor built specifically for neural-network inference, doing the matrix arithmetic at far lower power than a GPU. That silicon is the reason for the generational cut-off.

M1 and Above Only

Apple's own announcement lists the requirements:

  • Mac: M1 or later. Every Intel Mac excluded.
  • iPad: A17 Pro or M1 and later.
  • iPhone: iPhone 15 Pro / 15 Pro Max, and the iPhone 16 line (later joined by iPhone 16e).

The iPhone line was drawn sharply. The plain iPhone 15, still on sale at the time, was excluded; a customer who had bought one the previous year could not use the new Siri at all. On the Mac, every Intel machine dropped out — the hardware still ran Sequoia, but had no qualifying neural engine.

The same pattern Microsoft enacted with Copilot+ PCs and the Windows 11 TPM 2.0 debate repeated inside Apple's ecosystem: AI-feature generational gaps frozen as hardware-generational gaps.

The Intel exit continued afterwards. At WWDC in June 2025 Apple said that macOS 26 Tahoe, released that September, would be the last major macOS release to support Intel Macs at all.

The EU — It Was iPhone and iPad That Waited, Not the Mac

On 21 June 2024, days after WWDC, Apple issued a statement: because of the regulatory uncertainties introduced by the Digital Markets Act, it did not believe it would be able to roll out three features — iPhone Mirroring, SharePlay Screen Sharing enhancements, and Apple Intelligence — to EU users that year.

But that constraint bound iOS and iPadOS, not macOS. EU Mac users had Apple Intelligence from macOS 15.1 in October 2024, in U.S. English, on the same schedule as everyone else. EU iPhones and iPads got it with iOS 18.4 at the end of March 2025 — more than five months later.

The friction outlasted the rollout. On 24 September 2025 Apple published a statement arguing that the DMA requires it to make features work on non-Apple products before shipping them to its own users, and that the engineering this demands has delayed features in the EU. It named Live Translation with AirPods, iPhone Mirroring, and the Visited Places and Preferred Routes features in Maps. Of iPhone Mirroring it wrote: "Our teams still have not found a secure way to bring this feature to non-Apple devices without putting all the data on a user's iPhone at risk."

Siri 2.0 — Two Years Late, and a Different Thing

The most-watched part of the WWDC 2024 keynote was the next-generation Siri: personal context, on-screen awareness, actions taken across apps.

On 7 March 2025 Apple conceded the delay. Its spokesperson, Jacqueline Roy, said: "It's going to take us longer than we thought to deliver on these features and we anticipate rolling them out in the coming year." Apple did not say the features had missed a quality bar; it said they would take longer.

They took longer than that. By June 2025 the target was spring 2026; what actually materialised came at WWDC in June 2026 — and it was not built from Apple's own models alone. The rebuilt Siri runs on Google's Gemini, hosted on Apple's Private Cloud Compute. Bloomberg's Mark Gurman had reported in November 2025 that Apple was finalising a deal to pay Google roughly $1 billion a year for a custom 1.2-trillion-parameter Gemini model to power it. A company that spent a decade insisting it would do AI on its own terms bought the core of its flagship feature from a rival.

The gap between the marketing and the shipping reached court. In Landsheft v. Apple (N.D. Cal.), brought by buyers who said they had purchased iPhone 16 and iPhone 15 Pro models for Apple Intelligence's promised Siri, Apple agreed to a proposed $250 million settlement covering purchases between 10 June 2024 and 29 March 2025 — roughly $25 per device, up to $95 depending on claim volume. Apple denies wrongdoing, and final approval had not been granted as of this writing.

The ChatGPT integration did land on schedule, in 15.2 / iOS 18.2 in December 2024. But it was Apple borrowing OpenAI's model rather than shipping its own — and the 2026 Gemini deal is the same move at larger scale.

iPhone Mirroring and Deeper Continuity

A quieter but well-received feature was iPhone Mirroring — operating the iPhone's screen, notifications and apps directly from the Mac, and one of the high points of Continuity, the idea that Mac, iPhone, iPad and Apple Watch behave as a single device. As above, it was still unavailable in the EU as of September 2025.

The First Release Where Apple Was Behind

The macOS Sequoia / Apple Intelligence generation is likely to be remembered as the first Apple OS release in which Apple was the catching-up side.

Google had been shipping Bard / Gemini and Tensor-chip Pixel AI features since 2023. Microsoft had integrated Copilot into Windows and Office in 2023. Samsung launched Galaxy AI on the Galaxy S24 in January 2024. Against that timeline, Apple's AI only started running in October 2024, and its centrepiece did not arrive until 2026.

What survived was the shape of the architecture — on-device plus Private Cloud Compute, an attempt to guarantee privacy at the design level. Even the 2026 Siri, whose model is Google's, executes on Apple's own servers. Apple lost on speed, conceded on the model, and held on to the runtime. That is the compact summary of this generation.

Sources

  1. TertiaryApple Intelligence — Wikipedia

    Accessed 2026-08-08

  2. TertiarymacOS Sequoia — Wikipedia

    Accessed 2026-08-08

Last updated:

Share