Apple unveiled MacOS 26 tonight this night Tahoean update that marks the last stage of the transition to its homemade chips. If the new interface Liquid Glass fueling the debates on the user side, it is above all the integration of Apple Intelligence Who drew our attention with all the features of the announced AI, whether it be generation of images, summaries of notifications, writing tools or live translation, are reserved for Macs equipped with M1 and following Mac processors, Intel models, even recent, are excluded.
Faced with Microsoft, which relies on Copilot and integration into the Azure Cloud, and Google, which broadcasts Gemini through its online services, Apple chooses the ground of silicon. Since the acquisition of Palo Alto Semiconductor in 2008, the Cupertino firm has led to design its own chips, in order to get rid of Intel and carry out its AI functions directly on the motherboard of its devices. This choice allows it to display better performance, strengthen its argument around data confidentiality and consolidate user dependence on its ecosystem.
Apple focuses on the full control of the chain, from the design of the chip to applications, where its competitors favor the power of externalized calculation.
The fact remains that one of Apple’s main shortcomings today is the absence of a large reference language model capable of competing in quality with GPT-4, Claude 3 or Gemini. Models integrated in Apple Intelligence Are optimized to turn locally on M-Series chips, with low energy consumption and better data protection, but their size and power remain limited. This orientation makes it possible to offer practical and rapid functions (summaries, corrections, generation of simple images) but not yet to equal the versatility and the depth of the leading models of the market.
If Apple wants to go beyond the image of a “useful but modest” AI, it will have to invest in the development or acquisition of a large -scale LLM, whether internally trained (such as the Ajax project) or from a strategic partnership. Without this reference base, Apple risks being confined to a peripheral and defensive AI, when Microsoft and Google capitalize on their models to impose new standards of use.
