Version 5

Apple announced 'Siri AI' at WWDC 2026 on June 8, 2026, powered by a two-tier Apple Foundation Models architecture: lighter tasks on Apple's own hardware and cloud, heaviest workloads on AFM Cloud Pro running on Google Cloud with NVIDIA Blackwell GPUs [4][7]. iOS 27 will let users set Claude, ChatGPT, Gemini, or Grok as their default AI, and Apple has announced plans to open Siri to third-party AI tools more broadly [10][11][12]. A dispute over whether AFM Cloud is Apple's own model built with Google's help or a direct Gemini deployment remains unresolved [13][8][7]. Apple's conservative WWDC demo style was shaped in part by a $250M false advertising settlement over prior AI marketing claims [3].

Apple has structurally tied its largest AI workloads to Google's infrastructure and Gemini model technology due to a hardware scaling constraint rather than an optional partnership [8]. Whether Apple genuinely controls the end-to-end experience — versus depending on a competitor's model and cloud — is unresolved, and the $250M settlement context means Apple's careful demo posture reflects legal exposure, not just engineering caution.

Apple used WWDC 2026 on June 8, 2026, to relaunch its AI strategy under the 'Siri AI' name. The new assistant is designed as an operating layer that understands screen content, files, and applications, enabling actions across all installed apps in a single session [1]. Apple executives framed the approach as 'human-centered,' contrasting it with competitors they described as 'racing forward pursuing AI for the sake of AI' [2]. The keynote was pre-filmed with deliberate demo pauses. TechCrunch reported this presentation style was shaped partly by a $250M false advertising settlement Apple reached over prior AI marketing claims [3], which reframes Apple's conservative demo posture as legally motivated rather than purely engineering-cautious.

The technical architecture has two tiers. Lighter on-device and cloud tasks run on Apple Foundation Models. The heaviest tasks run on AFM Cloud Pro, confirmed at WWDC to operate on Google Cloud with NVIDIA GPU infrastructure [4][5]. Apple states it retains full software-layer control through Private Cloud Compute (PCC) regardless of hardware host [6], and committed to publishing all PCC binaries for public inspection [5]. NVIDIA confirmed its Blackwell GPUs with Confidential Computing are integrated into Apple's PCC on Google Cloud [7] — Confidential Computing isolates workloads in trusted execution environments so no party, including infrastructure operators, can access user data. Ars Technica offered a structural explanation for the Google arrangement: Apple's own PCC hardware cannot scale to Siri AI's capacity requirements, making the dependency architectural rather than elective [8]. Security-focused coverage has focused on the PCC extension to third-party data centers as the lead privacy story [9]. On the integration front, iOS 27 will allow users to set Claude, ChatGPT, Gemini, or Grok as their default AI [10], and Apple has announced plans to open Siri more broadly to third-party AI tools and developers [11][12].

The factual dispute over the model's origin has not been resolved. Apple's post-WWDC clarification described AFM Cloud as Apple's own model trained with Google's assistance [13], and analyst Neil Shah similarly framed Apple as having built AFM from the ground up with Google and NVIDIA as partners [14]. Ars Technica reported that Apple confirmed Siri AI uses Google Gemini language models running on NVIDIA hardware in Google-operated data centers [8], and MacRumors characterized the architecture as 'built around Google Gemini' [15]. NVIDIA's blog post occupies a middle position: it describes Apple Foundation Models as 'built in collaboration with Google' and as leveraging 'Gemini family model technologies' [7] — language that acknowledges Gemini's role without specifying whether Apple runs Gemini directly or distilled it into its own weights.

Reception is mixed. The Neuron identifies developer adoption as the key execution risk — Siri AI's practical utility outside Apple's own apps depends entirely on third-party developers building for the new AI layer [1]; Apple has announced that openness architecturally, but adoption has not yet materialized [11][12]. On social media, many observers read the launch as Apple delivering on promises made at WWDC 2024 two years late [16][17]. Simon Willison acknowledges the architecture looks more credible than Apple's 2024 effort but reserves judgment until features ship [5]. Apple's stock fell after the keynote despite entering WWDC near all-time highs [18].