Here’s how the Gemini-powered Siri will likely work under the hood

Google and Apple are reportedly nearing a $1 billion annual agreement for Google's Gemini model to power a revamped Siri, utilizing a 1.2 trillion parameter model hosted on Apple's Private Cloud Compute for enhanced privacy. This massive model is expected to leverage a Mixture of Experts (MoE) architecture to ensure efficient operation by activating only relevant sub-networks, thereby managing computational costs. While MoE is crucial for running such a large model efficiently, its long-term competitiveness against future AI advancements remains a key consideration for investors.

Analysis

Google and Apple are reportedly nearing a significant $1 billion annual agreement for Google's Gemini model to power a revamped Siri, slated for release next year. This collaboration involves a substantial 1.2 trillion parameter model, which will be hosted on Apple's Private Cloud Compute servers, a critical detail ensuring user privacy by preventing Google from accessing the data. This strategic partnership underscores both companies' commitment to advancing AI capabilities within their respective ecosystems.

The immense scale of the 1.2 trillion parameter model strongly suggests the implementation of a Mixture of Experts (MoE) architecture for efficient operation. MoE allows the model to activate only a fraction of its parameters, estimated at 75-150 billion, for each input, significantly reducing computational costs while leveraging a massive total parameter count. Apple already utilizes a form of MoE in its current cloud-based model, rumored to be 150 billion parameters, indicating a familiar approach.

Despite the large parameter count, the article highlights the challenge in directly comparing this model to other frontier AI offerings like GPT-5 or Claude Sonnet 4.5, whose parameter counts are undisclosed. A key consideration for investors is whether this 1.2 trillion parameter, MoE-driven model will remain competitive against rapidly evolving AI advancements by its projected launch next year. The overall sentiment is mildly positive, with positive sentiment specifically for GOOGL, GOOG, and AAPL, reflecting market optimism around the potential collaboration.

Key Decisions for Investors

• Monitor the finalization of the reported $1 billion Google-Apple AI partnership, as its confirmation could solidify strategic positioning for both companies in the generative AI space.
• Evaluate the long-term competitive viability of the 1.2 trillion parameter MoE model against rapidly evolving AI advancements, considering the undisclosed parameter counts of rival frontier models and potential for future disruption.
• Assess the market's reception to the privacy-centric approach of hosting the Gemini model on Apple's Private Cloud Compute, as this could influence user adoption and differentiate the offering in a privacy-conscious market.