Two of technology's largest companies moved to dominate consumer-facing AI this week, as Apple unveiled its third-generation Apple Foundation Models (AFM 3) and Google DeepMind launched Gemini Omni — a new multimodal model family designed to generate and edit video from a wide range of input types. The dual announcements signal a deliberate strategic positioning: as OpenAI has increasingly focused on enterprise customers and high-value API contracts, Apple and Google are racing to establish AI as a standard feature of the consumer devices billions of people already own.

Apple Foundation Models 3: Built With Google

AFM 3 represents Apple's most ambitious AI model release to date, and its first built in explicit collaboration with Google. The family comprises five models spanning on-device and cloud deployments: AFM 3 Core (smallest, on-device), AFM 3 Core Advanced (on-device with multimodal capabilities), and AFM 3 Cloud Pro, which runs on NVIDIA GPUs within Google Cloud infrastructure.

The Google collaboration is strategically significant. It reflects Apple's acknowledgment that Google's cloud AI infrastructure — built around its own TPU chips and optimised for large-scale inference — is more efficient for server-side workloads than Apple building equivalent infrastructure from scratch. For Google, hosting Apple's cloud models on Google Cloud is both a significant revenue contract and a demonstration of Google Cloud's competitive standing against AWS and Azure for AI inference workloads.

Apple Intelligence features built on AFM 3 include extended on-device reasoning, multimodal understanding of images and documents within iOS applications, and an improved Siri capable of executing multi-step tasks across third-party apps without requiring cloud connectivity for basic operations. The on-device focus remains central to Apple's privacy differentiation — user data for most AFM 3 Core operations does not leave the device.

Google Gemini Omni: Video Generation and Editing

Google DeepMind's Gemini Omni is the company's first model family specifically architected for video generation and editing alongside text, image, and audio capabilities. The Omni models are rolling out across the Gemini app, Google Flow (the company's creative AI workspace), and YouTube Shorts — where AI-assisted video editing features will be available to creators.

The specific video capabilities include: generating short video clips from text prompts, editing existing video by describing changes in natural language, extending clips with AI-generated continuation, and combining multiple input types (images, music, narration) into coherent video output. Early demonstrations showed particular strength in stylistic consistency and temporal coherence — problems that plagued earlier video generation models.

Gemini Omni Flash, the faster and lower-cost tier, began rolling out immediately. The full Omni model is expected to reach broader availability in August 2026.

The OpenAI Context

OpenAI's enterprise pivot — evident in GPT-5.6's $200/month Pro tier and the company's deepening relationships with government and Fortune 500 customers — has created space for Apple and Google to compete on consumer AI without directly confronting OpenAI on its strongest ground. ChatGPT remains the largest consumer AI product by active users, but the integration of AFM 3 into every iPhone and Gemini into every Android device and Google product could shift that balance significantly over the next 12–18 months.

The key consumer AI battleground is the personal assistant layer: which AI has the most contextual awareness of a user's apps, documents, communications, and preferences. Apple's on-device advantage and Google's search and productivity data give both companies structural advantages that API-first companies cannot easily replicate.

What This Means for Users

For iPhone users running iOS 20 (expected September 2026), AFM 3 delivers improved on-device writing assistance, summarisation, and image understanding without requiring a subscription or cloud connection for most tasks. For Android users and Google Workspace subscribers, Gemini Omni enables video content creation that previously required dedicated creative software or professional production budgets. Both announcements represent a genuine capability expansion that reaches consumers through products they already use daily — the most powerful distribution advantage in the AI market.

Developer Implications of the Apple-Google Collaboration

For third-party developers, the Apple-Google AI collaboration raises significant questions about the future of the iOS and Android ecosystems. Apple has historically kept its AI infrastructure proprietary, making the Google partnership a notable departure. The implication is that Apple is prioritising capability delivery over infrastructure self-sufficiency — a pragmatic choice that accelerates iOS AI features but introduces a dependency on Google's cloud that Apple's privacy positioning must work to accommodate.

Apple's privacy engineering team has disclosed that AFM 3 Cloud Pro requests are processed under what Apple calls "Private Cloud Compute" architecture — a system in which user data is processed on servers specifically isolated for the request and not retained for model training. The architecture is designed to provide the same privacy guarantees as on-device processing while enabling the scale and capability of cloud inference. Independent security researchers are expected to audit these claims as part of Apple's bug bounty programme.

For developers building on iOS, AFM 3 opens new API surface through Apple Intelligence Extensions — the framework that allows third-party apps to surface AI capabilities within iOS. Apps that integrate the Extensions framework gain access to AFM 3's multimodal understanding, writing assistance, and task execution capabilities without building their own model infrastructure. This positions the App Store as a distribution layer for AI-enhanced applications in a way that no third-party AI API can replicate at scale.

Google Gemini Omni: Content Creator Use Cases

The most immediate practical impact of Gemini Omni is on solo content creators — YouTubers, social media managers, and small marketing teams — who currently lack access to professional video production tools. The ability to describe a video edit in natural language and have it executed automatically reduces the barrier to producing polished video content significantly.

Early use cases demonstrated during Google's preview include: converting long-form interview recordings into short clips with auto-generated captions and B-roll, creating product demonstration videos from static product images, extending short clips with AI-generated continuation that matches the original style and lighting, and generating animated social graphics from text descriptions.

The monetisation path for Google is clear: Gemini Omni capabilities inside YouTube Shorts encourage creators to produce more content on the platform, increasing supply of short-form video and driving ad revenue. YouTube's ad revenue — approximately $10 billion per quarter in recent periods — is the primary financial return on Google's investment in Gemini Omni's creator tools.

The Privacy and Regulatory Frontier

Both AFM 3 and Gemini Omni will face scrutiny under the EU AI Act's provisions on general-purpose AI systems. Models with compute above the regulatory threshold — which both almost certainly exceed — are required to submit technical documentation, comply with copyright rules for training data, and publish model cards. Neither Apple nor Google has disclosed full training data provenance for the new models, which will be a point of focus for EU regulators.

California's AI transparency law, passed in March 2026, imposes additional disclosure requirements for models deployed to California residents — the largest state-level AI regulation in the US to date. The convergence of EU, UK, and California-level regulation means that the largest AI deployments now face a patchwork of compliance requirements even before any federal US AI framework is enacted.