Apple Is in Talks to Let Google's Gemini Power iPhone Generative AI Features

Hardware, Scale, and Infrastructure

• Several argue neither Google nor OpenAI can fully handle Apple-scale demand yet, though others note Google’s TPU-based Gemini stack is massive and still expanding.
• Debate over TPUs vs GPUs: TPUs seen as highly efficient for inference, but less popular and harder to use outside Google; Nvidia GPUs favored for training due to CUDA, ecosystem, and broader availability.
• Some expect Apple to offload as much as possible to users’ devices via on‑device models, easing cloud load; others doubt current phone hardware can support GPT‑4‑class capabilities locally.

On‑Device vs Cloud AI

• Many expected Apple to lean heavily on efficient on‑device inference and are disappointed by a reliance on an external cloud model.
• Others read this as a hybrid “escape hatch”: small, private models locally, heavier generative tasks via Gemini until Apple’s own stack matures.
• Concerns raised about battery drain and economics of local general‑purpose AI; several predict cloud will remain dominant for heavy tasks.

Privacy and Data Concerns

• A major worry is Apple sending user data to Google, seen as contradicting Apple’s privacy branding and a key reason some choose iOS.
• Counterpoint: Apple already routes huge volumes of searches to Google and both platforms transmit substantial encrypted telemetry.
• Some discuss difficulty of inspecting device traffic (certificate pinning, closed ecosystems), arguing users can’t fully verify privacy claims.

Business Strategy and Competence

• Some interpret the move as evidence Apple is behind in AI despite acquisitions and resources, even “unambitious” compared to smaller players.
• Others see it as a standard buy‑vs‑build decision, a stopgap, or a way to outsource legal and reputational risk around training data and model outputs.
• There’s concern that outsourcing core AI could weaken Apple long‑term while strengthening Google’s position in mobile AI.

User Experience and Assistants

• Frustration with Siri’s stagnation and regressions is common; many hope Gemini-level capabilities will finally make assistants genuinely useful.
• Some envision assistants becoming primary OS interfaces, others find that prospect frustrating and favor GUI or even brain–computer interfaces.