Report: 90% of Apple's Siri Servers Are Idle
A new report claims Apple's server-side AI infrastructure for Siri is chronically underutilized, with up to 90% of its capacity sitting idle. The analysis points to a significant mismatch between technical investment and business strategy, especially as cloud-native competitors like Google's Gemini gain ground. The situation highlights the risks of over-provisioning and a lack of cross-team coordination between infrastructure and product teams.
The underutilization of Apple's Private Cloud Compute (PCC) is reportedly a consequence of fragmented internal AI infrastructure. Different teams have historically operated in silos, using varied technology stacks, which has prevented the creation of a unified, efficient server architecture. This has led to redundant infrastructure and cost overruns, with multiple attempts to centralize the system stalling over the last decade. This situation has led to a deeper reliance on third-party providers, specifically Google. Apple is reportedly in discussions to have Google host the revamped, Gemini-powered Siri within its data centers, albeit while adhering to Apple's privacy standards. This move is seen as a pragmatic, short-term solution to bypass internal infrastructure bottlenecks and meet market demands for advanced AI features. The server hardware itself, based on modified M2 Ultra processors, is described as underpowered for running the latest large-scale AI models. This has resulted in some manufactured servers remaining unused in warehouses. Concurrently, Apple is developing a custom AI server chip, codenamed "Baltra," with an expected deployment in 2027-2028, indicating a long-term strategy to bring this critical infrastructure in-house. Apple's overarching AI strategy continues to emphasize on-device processing for privacy and speed, leveraging the powerful Neural Engine in its custom silicon. The PCC is intended to handle more complex requests that can't be processed on-device, functioning as a secure extension of the user's hardware. This hybrid approach aims to balance performance with Apple's stringent privacy principles, ensuring user data is not stored or made accessible to Apple. Recent leadership changes reflect a strategic realignment of the AI division. John Giannandrea, who joined from Google in 2018, is transitioning to an advisory role, with Amar Subramanya, another former Google and Microsoft executive, taking over as VP of AI. Subramanya now reports to software engineering chief Craig Federighi, creating a more unified structure for developing Apple's foundational models and integrating AI features across all platforms.
