Apple's AI strategy undergoes a major transformation! Self-developed data center chips make a foray into the cloud, supporting AI features in iOS 18 this fall
Media reports that Apple, which has long prioritized AI on the client side, will use its high-end self-developed chips for Mac configurations in cloud computing servers. This will allow Apple's devices to perform advanced AI tasks, while relatively simple AI functions will be directly implemented by the internal chips of iPhone, iPad, and Mac; the first server chip is the M2 Ultra
Author: Li Dan
Source: Hard AI
Last week, Apple CEO Tim Cook hinted at the upcoming Worldwide Developers Conference (WWDC) in the next month, announcing "significant AI plans." This week, the media exposed Apple's roadmap for a major advancement in artificial intelligence (AI) functions this year: using its own data center chips to fully launch AI functions on the edge and in the cloud.
On Thursday, May 9th, Eastern Time, media reports cited sources saying that Apple will launch AI functions through its self-developed data center chips this year. They plan to use self-developed chips similar to those in Mac computers for cloud computing servers, allowing Apple's devices to perform cutting-edge AI tasks, such as generating images, summarizing lengthy articles, creating long-form email responses, and upgrading the Siri voice assistant.
Simpler AI functions will directly land on iPhones, iPads, and Macs, implemented by the chips inside these Apple devices, such as summarizing missed iPhone notifications or received messages for users.
The media mentioned that Apple has been brewing the use of self-developed chips for cloud-based AI tasks for about three years. After the AI application craze sparked by ChatGPT and Gemini, Apple was forced to accelerate the implementation of this AI plan. Apple's first server chip will be the M2 Ultra, and Apple is already considering future server chip versions based on the M4 chip.
When the M2 Ultra was launched in June last year, Apple claimed it to be the largest and most powerful chip to date, empowering the new Mac Studio and Mac Pro to become the most powerful Mac computers to date. The M2 Ultra uses a second-generation 5-nanometer process, connecting two M2 Max chips, integrating a total of 134 billion transistors internally, 200 billion more than the M1 Ultra; the maximum memory capacity can reach 192GB, 50% higher than the M1 Ultra, with a memory bandwidth of up to 800GB/s, twice that of the M2 Max.
The central processing unit speed of the M2 Ultra is up to 20% faster than the M1 Ultra, its graphics processor speed is up to 30% faster, and the neural network engine speed is up to 40% faster. Compared to the M2 Max, the media processing engine performance speed of the M2 Ultra is up to twice as fast.
The media mentioned that the new AI functions will be part of the iOS 18 system released by Apple this fall, and currently, Apple plans to run cloud functions using its own data centers, but will eventually rely on external infrastructure In other words, data centers equipped with Apple chips will support the AI functions of iOS 18.
The media believes that this approach represents a shift for Apple. For years, Apple has always prioritized on-device AI, seeing it as a better way to ensure security and privacy. Those involved in creating the Apple Service Center chip project codenamed ACDC revealed that existing components in the processor can protect user privacy. Through a method called Secure Enclave, Apple can isolate data from security vulnerabilities.
Last week, Wall Street News mentioned that some media outlets reported that Apple is developing chips for data center servers, with the internal project codenamed ACDC. This chip may focus on running AI models, known as inference, rather than training AI models, which may still be dominated by products from NVIDIA. Some media outlets also mentioned that in terms of AI software, Apple will focus on providing new proactive features for users in their daily lives.
Regarding the news that Apple will introduce AI functions with data center chips, some netizens asked tech industry reporter Mark Gurman, who leaked the information, whether this represents a 180-degree turnaround in his attitude the next day. Gurman said yes, as he previously replied that he did not believe Apple was committed to AI server chips, but now he thinks Apple will use existing high-end Mac chips for servers.
Some netizens commented that it feels like Apple's philosophical thinking has undergone a major shift from on-device AI to data center AI, and they are curious about what it will be used for. Some netizens are eager to see how well Apple's user experience combined with AI can be. Another netizen mentioned the AI chip leader NVIDIA, saying that this news from Apple does not mean NVIDIA is finished