The Zhitong Finance App learned that Apple (AAPL.US) is currently using Amazon (AMZN.US) AWS's customized artificial intelligence chips for search and other services, and will evaluate whether the company's latest artificial intelligence chips can be used to pre-train models such as Apple Intelligence.

At the annual AWS re:Invent conference held on Tuesday, Apple revealed its use of Amazon's proprietary chips. Benoit Dupin, Apple's senior director of machine learning and artificial intelligence, took the stage to discuss how Apple uses cloud services.

“We have strong relationships, and our infrastructure is both reliable and able to serve customers around the world,” Apple's Dupin said.

Apple's appearance at the Amazon conference and its acceptance of the company's chips is certainly a strong support for this cloud service provider, as Amazon is competing with Microsoft (MSFT.US) Azure and Google (GOOGL.US) Google Cloud for artificial intelligence spending. Apple also uses these cloud services.

Dupin said that for more than 10 years, Apple has been using AWS to provide services such as Siri, Apple Maps, and Apple Music. For example, Apple uses Amazon's Inferencia and Graviton chips to provide search services. Dupin said Amazon's chips have increased efficiency by 40%.

Dupin also said that Apple will use Amazon's Trainium2 chip to pre-train its proprietary models. This shows that compared to x86 central processors manufactured by Intel (INTC.US) and AMD (AMD.US), Amazon's chips are not only a cost-effective way to infer artificial intelligence models, but can also be used to develop new artificial intelligence. Amazon announced on Tuesday that its Trainium2 chip is fully open for rent.

“In the early stages of evaluating Trainium 2, we expected to increase efficiency by 50% through pre-training,” Dupin said.

AWS CEO Matt Garman said in an interview on Tuesday that Apple has always been an early adopter and beta tester of the company's Trainium chips.

Garman said Apple “came to us and said, 'How can you help us achieve generative AI capabilities, we need infrastructure to build', and they have this vision for building Apple Intelligence.”

Earlier this year, Apple said in a research report that it had used Google Cloud's TPU chip to train its iPhone artificial intelligence service, which Apple called Apple Intelligence.

Currently, most artificial intelligence training on the market is done on expensive Nvidia (NVDA.US) GPUs. Cloud providers and startups are racing to develop alternatives to lower costs and are exploring different ways that may drive more efficient processing. Apple's use of custom chips may signal to other companies that non-Nvidia training programs can also work.

This fall, Apple unveiled its first major generative artificial intelligence product. Apple Intelligence is a series of services that can summarize notifications, rewrite emails, and generate new emojis. The company said it will integrate with OpenAI's ChatGPT later this month, and next year, Siri will get new features to control apps and speak naturally.

Unlike leading chatbots such as OpenAI's ChatGPT, Apple's artificial intelligence solutions are not based on large Nvidia-based cloud server clusters. Instead, Apple uses iPhone, iPad, or Mac chips to complete as much processing as possible, and then uses its own M-series chips to send complex queries to servers operated by Apple.