Apple's approach to AI hardware is currently encountering substantial hurdles, as evidenced by the notably low utilization rates of Apple Intelligence features. The company's internal infrastructure is fragmented, resulting in a situation where up to 90% of its private cloud computing servers remain idle, with merely 10% of the available computing power being actively utilized. This challenging scenario arises from the autonomous functioning of Apple's various R&D teams, each employing disparate technology stacks and failing to establish a cohesive server technology architecture. Consequently, this has led to inefficiencies and escalated costs.

To mitigate this issue, Apple has provisionally resorted to using Google servers to fuel the next-generation Siri chatbot. The company has plans to roll out a substantially overhauled version of Siri within this year, which will be underpinned by the computational prowess of a tailored Gemini AI model. Looking ahead to next year, with the advent of the iOS 27 update, Apple intends to introduce a standalone Siri chatbot. This new iteration will operate on a system that leverages Google TPUs and cloud infrastructure.

To tackle the problem at its core, Apple is making strides in its in-house AI server chip project, codenamed 'Baltra.' This endeavor is being undertaken in collaboration with Broadcom, with the chip being manufactured utilizing TSMC's cutting-edge 3nm 'N3E' process. The chip is set to incorporate a chiplet architecture, marking a significant step forward in Apple's quest to optimize its AI hardware strategy.