Apple appears to be facing severe internal infrastructure pains in its pursuit of generative AI. (According to The Information website...)Get the messageBecause the actual usage of Apple Intelligence is lower than expected, coupled with the severe fragmentation of the technical architecture of its internal R&D teams, up to 90% of the AI server computing power Apple has prepared for its Private Cloud Computing (PCC) is actually idle. To address this immediate crisis, Apple has previously confirmed that it will expand its cooperation with Google and directly introduce Gemini technology to enhance the new version of Siri that will be launched this year.
The interior is dotted with numerous smaller hills: 90% of the computing power of private cloud servers is idle.
According to the data disclosed in the report, of the hardware computing power that Apple has reserved to support Apple Intelligence cloud computing, only about 10% has actually been deployed to data centers, and the remaining 90% of AI servers are currently idle.
The underlying reason for this severe waste of resources lies in the "severe fragmentation" of Apple's internal AI infrastructure. Apple's different AI R&D teams are operating independently, using drastically different technology stacks, and have been unable to establish a unified and efficient server technology architecture.
This fragmented R&D model not only dragged down overall development efficiency, but also caused serious cost overruns due to redundant infrastructure construction.
Short-term compromise: Partnering with Google to power the new version of Siri with Gemini technology
Faced with internal infrastructure bottlenecks and the market's urgent desire for AI applications, Apple chose a pragmatic short-term compromise: to deepen its cooperation with Google and use the Gemini model as the core engine for upgrading Siri.
This year's major overhaul: Apple is expected to release a significantly redesigned Siri this year. This new version will be powered by a custom Gemini AI model (internal codenamed Foundation Models v10) with up to 1.2 trillion parameters, comprehensively enhancing Siri's understanding and conversational abilities.
A Standalone Chatbot in iOS 27 Next Year: Looking to the future, Apple plans to introduce a more powerful standalone Siri chatbot in next year's iOS 27 update. Although the system is owned by Apple, it will run on Google's TPU and cloud infrastructure, and will feature a more advanced Foundation Models v11 with performance expected to rival that of the Gemini 3.
Long-term strategy: Self-developed AI server chip codenamed "Baltra"
Although Apple will have to rely on Google's support in the short term, it clearly has no intention of handing over the lifeline of cloud AI to a competitor in the long run.
To fundamentally reverse its dependence on foreign technology, Apple is pushing forward with a self-developed AI server-specific integrated circuit (ASIC) codenamed "Baltra." This chip is expected to feature the following advanced specifications:
• A powerful alliance:We have entered into a deep design collaboration with Broadcom, a major manufacturer of networking chips.
• Advanced process:It is manufactured using TSMC's current top-of-the-line 3nm "N3E" process.
• Chiplet architecture: By integrating different functional modules within a single package, the Chiplet design not only significantly optimizes communication efficiency within the chip, but also allows Apple to maintain strict confidentiality regarding the overall chip architecture design details to its partners.
Analysis of viewpoints
This leak reveals Apple's two-pronged approach to its AI strategy: "Front-end experience relies on alliances, while back-end infrastructure relies on in-house development."
The fact that up to 90% of its private cloud computing power is idle does expose Apple's shortcomings in cloud software and server architecture management. However, Apple's brilliance lies in its willingness to lower its standards and, during a potential downturn in user experience, directly introduce Google's industry-leading Gemini technology to "extend the life" of Siri. This ensures that hundreds of millions of iPhone users can enjoy a first-class AI voice assistant experience this year, without being lost due to Apple's internal technical growing pains.
Meanwhile, Apple's real trump card is its self-developed chip project codenamed Baltra. Once this server chip, based on TSMC's 3nm process and Chiplet architecture, is completed, Apple will be able to perfectly replicate the energy efficiency advantages it has accumulated with Apple Silicon (M-series chips) in its data centers. At that time, Apple will not only be able to completely break free from its dependence on Google Cloud, but also regain absolute control over hardware and software integration.
