Apple to ditch Nvidia? New AI system rumored to collaborate with Google

Apple's technical paper on Monday revealed that two artificial intelligence models of Apple Intelligence were pre-trained on Google's custom cloud chips. One is called TPU (Tensor Processing Units) from Google.

Google's TPU Tensor Processing Units were introduced in 2015 and publicly released in 2017. Originally created for internal workloads, TPUs have now found broader applications. Apple's move validates that top tech companies in the field of training AI models are looking for alternatives to Nvidia GPUs.

Nvidia's expensive Graphics Processing Units (GPUs) dominate the high-end AI training chip market, with demand soaring in recent years to the point of shortages. Open AI, Microsoft, Anthropic all use Nvidia's GPUs in their models, while other tech giants including Google, Meta, Oracle, and Tesla are also buying Nvidia chips to build their AI systems and products.

Reuters noted that while Apple did not explicitly state that they did not use any Nvidia chips, the company was not mentioned in their description. In its 47-page technical paper, Apple did not mention Google or Nvidia, but did indicate that their Apple Foundation Models (AFM) and AFM servers were trained on Google's cloud TPU clusters. This indicates that Apple rented servers from a cloud provider for computation. Apple stated in the article: The system allowed us to efficiently scale training of AFM models, including on-device AFM, AFM servers, and larger models.

Apple Uses Google Chips to Train Two AI Models

Reuters reported that Apple's disclosures show it will rely on Google chips, not the industry-leading Nvidia-designed graphic chips, to build the artificial intelligence infrastructure for its upcoming Apple Intelligence.

Apple's reliance on Google's cloud infrastructure is notable as Nvidia holds about 80% market share, producing the most popular AI graphic processors. Chips produced by Google, Amazon, and other cloud computing companies also require collaboration with Nvidia.

In the research report, Apple did not explicitly state that it does not use Nvidia chips, but its description of AI hardware and software did not mention Nvidia's hardware.

The iPhone manufacturer used two forms of Google Tensor Processing Units (TPUs) to train AI models, which were organized in large chip clusters. Apple used 2,048 TPUv5p chips to create AI models that can run on iPhones and other devices. Apple deployed 8,192 TPUv4 processors for its server AI models.

Nvidia focuses on GPUs widely used for AI work. Google and Nvidia differ in how they sell chips and systems as standalone products; Google sells access to TPUs through the Google Cloud Platform, and customers interested in purchasing access must build software through Google's cloud platform system to use these chips.

Apple has rolled out some Apple Intelligence features to its beta users. Reuters reported the use of TPU chips in June, but Apple's recent research revealed the extent of its full reliance on Google hardware.

Apple engineers stated in the report that using Google's chips can produce larger and more complex models than the two models studied in the report. Apple introduced a range of new AI features at its developer conference in June, including integrating OpenAI's ChatGPT technology into its software.

Investors Concerned about Future Applications of Apple Intelligence on iPhone

Apple introduced Apple Intelligence at its 2024 developer conference, integrating personal intelligence systems in iOS 18, iPad OS 18, and macOS Sequoia.

Apple Intelligence consists of multiple generative models for users' daily tasks, dynamically learning to perform their current tasks. Built-in base models fine-tune users' actions such as writing and polishing text, prioritizing and summarizing notifications, creating engaging images for conversations with family and friends, and simplifying built-in app features to reduce app switching.

Recent iPhone models and Siri performance by Apple have been subpar, and the newly introduced Apple Intelligence is only available for iOS 18, requiring at least an iPhone 15 Pro or higher, MacOS 15 and iPadOS 17 to use.

Experts noted from a Bloomberg report that Apple's delay in launching Apple Intelligence until the end of the year, missing the upcoming iPhone firmware update in September, means developers can only start testing new AI features this week at the earliest. Investors and Apple enthusiasts are looking forward to the new Apple iPhone 16 at the end of the year with Apple Intelligence AI enhancements, which may lead to disappointment.

Apple's slow pace in hardware and software application development, and its inseparable relationship with Google, raises the question of whether Apple Intelligence will be a standout product to replace Siri or just a revamped Siri 2.0. The iPhone, once a groundbreaking product introduced by Jobs, is no longer the only choice for smartphone users in the rapid advancement of AI. Apple's new AI features will undoubtedly impact its position and stock price in the tech industry.