Two tiers, one label
At WWDC 2026, Apple revealed an important distinction within the Apple Intelligence architecture: alongside the established on-device model, there is now a cloud tier called AFM Cloud Pro. According to CNBC, this model targets "the most demanding tasks" and is "similar in quality to Gemini Frontier models." The most significant technical detail is its infrastructure: Nvidia GPUs running in Google's cloud, a direct result of the Apple-Google collaboration announced in January 2026.
Apple also presented an updated Foundation Models framework for developers. CNBC and Engadget confirm the model can now understand speech, read text, and interpret images as input. MacRumors details that the framework gains image input support, custom skills, and server-side model execution, meaningfully expanding what developers can build.
The infrastructure question
The fact that Apple's most capable model runs on Nvidia hardware in Google's cloud is worth noting plainly: Apple Intelligence's advanced inference capacity depends, at least for now, on an external supplier. Apple maintains a privacy narrative through Private Cloud Compute — Google does not retain data after processing — but computational sovereignty remains partial. For developers, access to a Gemini Frontier-quality model via Apple APIs still represents a meaningful step forward.