Audience members watching Apple’s keynote at WWDC 2025. Photo: Bloomberg .
Apple plans to showcase an AI model running on its devices at its developer conference in June in order to attract attention.
According to The Information , Apple’s upcoming WWDC 2026 conference will focus on its ability to run artificial intelligence directly on devices. The company could leverage 15 years of chip design experience to ensure seamless and stable functionality.
Specifically, Apple is expected to share the advantages of its iPhone, Apple Watch, and Mac processors in handling native AI queries. The company will also highlight the advantages of on-device AI compared to cloud-based AI.
According to MacRumors , citing The Information , “While the cloud will remain necessary for complex queries, Apple will prioritize on-premises processing, offering cost savings and better privacy protection compared to operating in massive data centers pursued by competitors.”
As part of its agreement with Google, Apple is expected to use a larger version of Gemini to train a smaller prototype that can run directly on the hardware.
Apple is also considering acquiring companies in this field to downsize its models. One of the companies it’s eyeing is Liquid AI, a Massachusetts-based startup specializing in developing locally operated AI models.
According to the statement, some queries still require cloud processing. Apple is said to have agreed to use Nvidia’s computing technology on Google Cloud to process larger models.
This system can encrypt data and AI models during processing, which may slightly increase performance costs but provides better security.
If the rumors are true, this deal is significantly different from when Apple Intelligence was first launched. Back then, Apple stated that all cloud queries were processed by a custom cloud computing server system powered by its own chips.
Despite running on Google’s servers, Apple is rumored to still use the name Private Cloud Computing.
Despite prioritizing local processing, rumors suggest Apple’s model still has significant limitations. The full version of Gemini contains trillions of parameters. Meanwhile, the company is reportedly having difficulty running this model on its own servers, which use the same processors as Mac computers.
Apple Intelligence was first introduced at WWDC 2024. However, its launch faced numerous obstacles due to low user adoption and delays in releasing various features, including a new version of Siri. The company plans to officially announce the new assistant on June 8th at WWDC 2026.
Source: https://znews.vn/apple-no-luc-vuot-kho-post1655113.html
