Apple’s Private Cloud Compute to run on Google Cloud

How the AFM Model Family Splits Workloads

Apple has partnered with Google to power the high-end reasoning capabilities of its Apple Intelligence suite, running its most demanding models on Google Cloud infrastructure. The collaboration integrates Google’s NVIDIA-powered servers with Apple’s Private Cloud Compute to deliver a more personalized Siri while maintaining strict user privacy guarantees.

How the AFM Model Family Splits Workloads

The backbone of this integration is the third generation of Apple Foundation Models (AFM), a family of five models developed in collaboration with Google. Apple has designed a tiered system that shifts processing between on-device hardware and remote servers based on the complexity of the user’s request.

Simple tasks are handled locally to maximize speed and privacy. For example, asking Siri for a recipe typically triggers on-device processing. However, complex reasoning—such as analyzing a 40-page PDF contract to identify risky clauses—is routed to the AFM 3 Cloud Pro model. This specific model does not reside on Apple’s own network but runs on NVIDIA GPUs within Google’s data centers.

How the AFM Model Family Splits Workloads
Photo: Apple Machine Learning Research
Model Name Location Primary Function Key Specification
AFM 3 Core On-Device Basic intelligence tasks 3-billion-parameter dense model
AFM 3 Core Advanced On-Device Multimodal tasks, dictation 20-billion-parameter sparse model
AFM 3 Cloud Server-side Speed and efficiency workhorse Private Cloud Compute
ADM 3 Cloud (Image) Server-side Image generation and editing Powers Image Playground
AFM 3 Cloud Pro Google Cloud Complex reasoning, agentic tools NVIDIA GPU powered

To manage memory constraints on consumer hardware, Apple researchers developed Instruction-Following Pruning (IFP). This allows the AFM 3 Core Advanced model to store its full weight in flash memory rather than active DRAM, activating only 1 to 4 billion parameters at a time depending on the prompt.

Siri AI’s New Capabilities and Hardware Requirements

The most visible result of this architecture is the transition to Siri AI. According to Engadget, the overhaul introduces a “more personalized Siri” that can take action based on screen content. Users can now swipe down from the middle of the screen in iOS 27 to bring up a chatbot interface where responses appear in cards.

Why Apple’s Private Siri AI Runs on Google Cloud

Visual Intelligence is a core pillar of the update. Siri AI can now analyze what a camera sees or what is currently displayed on a Mac or iPhone screen. Apple demoed the ability to look at a photo of a shed and ask the assistant how to build a maker space inside it, or to extract performances from a festival schedule screenshot and add them directly to a calendar.

  • iPhones: iPhone 16 models or later, and the iPhone 15 Pro and Pro Max.
  • iPads: iPad mini (A17 Pro) and iPad models with M1 chips or later.
  • Macs: Any Mac with an M1 chip or later.
  • Wearables: Apple Watch Series 10 or later, Ultra 2, and SE 3 (when paired with an Apple Intelligence iPhone).
  • Other: Apple Vision Pro.

The software will launch this fall across iOS 27, iPadOS 27, watchOS 27, visionOS 27, and macOS Golden Gate, initially supporting only English.

The Strategic Risk of Google Dependency

While the partnership solves Apple’s immediate generative AI deficit, it creates a precarious dependency. The Business Times reports that Apple is increasingly becoming a “wrapper” around Google’s capabilities, potentially handing Alphabet significant negotiating leverage.

The Strategic Risk of Google Dependency
Photo: The Business Times

This relationship is already financially intertwined. Sealed court documents from a US antitrust lawsuit revealed that Google paid Apple US$20 billion to remain the default search engine on iPhones. Now, Google’s cloud and AI infrastructure are essential for the very features Apple is using to modernize its ecosystem.

“Apple has made privacy central to its brand, but partnering on AI with a business founded on collecting personal data certainly muddies the waters.”

The Business Times

Apple attempts to mitigate this brand risk by insisting that no user data is stored on Google’s servers and that Google cannot use Apple’s data to train its own models. However, the operational reality remains that Apple’s most “capable” server-based model relies on a competitor’s data centers to function.

Regulatory Hurdles in the EU and China

Despite the global rollout plan, Apple Intelligence and Siri AI will not be available in the European Union or China at launch. The company cited regulatory reasons for these omissions, likely stemming from the EU’s Digital Markets Act and China’s strict AI governance laws.

This creates a fragmented user experience where some of the most advanced hardware—such as the iPhone 16—will lack its headline software features depending on the region. Apple has not provided a specific timeline for when these markets will receive the update, though it indicated that support for more languages will be added soon.

The next 30 days will focus on developer betas, allowing programmers to test how Siri AI integrates with third-party apps before the general public release this fall. The ultimate success of the project depends on whether users accept the trade-off of routing their most complex requests through Google’s cloud in exchange for a more capable virtual assistant.

Find more reporting in our Technology and Science section.

Related Posts

Leave a Comment