TheChek News Flow

The project would turn Apple into a rival of Amazon, Microsoft and Google

Apple over the past few years considered launching a business renting out its servers to the millions of developers that create apps for the iPhone and Mac, in what would be a hugely ambitious move that would pit it against Amazon, Microsoft and Google.

The idea discussed was for Apple to rent out servers running on Apple’s own chips, said three people involved in the effort. That reflects Apple’s successful development of its own chips, which have given its phones and laptops a big edge over rival devices—and which are also now being used in Apple’s data centers for some artificial intelligence services.

Apple executives think its chips perform so efficiently that developers would end up spending less on its cloud service than they do on major cloud firms now, especially for intensive AI applications. The status of the project is, however, unclear. The biggest proponent of the idea, a cloud executive called Michael Abbott, left the company in 2023. While discussions were still active as of the first half of 2024, it’s not known whether they have continued.

Apple declined to comment.

Even if Apple doesn’t go ahead with the service, the internal discussion demonstrates the strength the company has in semiconductor design, more than other tech companies that have designed their own chips—including Microsoft, Amazon and Google. On Wednesday, The Information reported that Microsoft was scaling back its in-house chip ambitions following delays.

Photos and Music

Apple has tested its chips for servers to process Apple Wallet transactions in the cloud, showing it could save money on internal infrastructure costs, the people said. Apple’s Photos and Apple Music apps have also tried using the chips and saw improved performance in areas like search, said people familiar with the effort.

While Apple has been struggling to keep up in AI, its chips have emerged as a potential bright spot for the company in powering more basic types of AI. It launched its first chip for its iPad and iPhone in 2010 and started adding specialized AI capabilities in 2017. Apple’s chips are efficient at running many types of inference, which means using pre-trained AI models to interpret new information, such as computer vision found in the Vision Pro.

Big tech companies and AI startups are spending tens of billions of dollars annually on Nvidia chips, which were originally designed for graphics processing in gaming, for the intensive process of training large AI models. The launch of ChatGPT in late 2022 has spurred even larger models that require more processing horsepower, mostly using Nvidia chips.

Apple has mostly resisted that push to spend heavily on Nvidia chips for AI training. Instead it has prioritized renting chips from Amazon and Google clouds, spending around $7 billion annually to do so, said people familiar with the company’s cloud spending. Only a small portion of that spending is for AI training, however.

The benefit of using outside clouds for training is that Apple won’t be stuck with a large inventory of Nvidia chips after they’ve trained the models.

But increasingly, as companies seek to make an actual business off the AI models they’ve spent billions training, inference will become a larger portion of AI computing. That’s where Apple thinks it will find an advantage over competitors.

“Apple has been doing AI on their chips for the longest that anyone has been doing it,” said Amit Kumar, CEO of Dragonfruit, a company that uses Apple chips to power a security camera monitoring service for retailers. “While Nvidia was doing gaming, Apple was doing actual AI compute.”

Dragonfruit is one of a growing set of companies turning Apple computers into something like a server. The company buys up Mac minis, a tiny puck-shaped computer without a screen or user interface like a keyboard, and strings them together to create AI machines to analyze camera data at stores. Kumar said that by going with Apple chips, his company gets a massive cost advantage over rivals that use Nvidia chips.

Another company, webAI, offers businesses the ability to efficiently run AI models across any Apple hardware. Apple is also building an open-source framework, MLX, for developers running AI on Apple chips. At the company’s most recent developer conference last month, Apple also opened up its so-called foundation models, allowing developers to run AI on-device within their apps.

Launching a cloud service for developers would help expand Apple’s services revenue, one of Apple’s last remaining areas of growth, at a time when that business is in danger. New laws and court orders threaten Apple’s lucrative App Store fee it collects, and the U.S. Justice Department is challenging the $20 billion default search deal it collects from Google every year.

Instead of having to rely on a large enterprise sales team like a traditional cloud firm, Apple would assign its developer relations team to handle developers who sign up for the service, the involved people said.

Still, launching a cloud service would also be an unusual move for a company that has been so steadfast in its commitment to consumer businesses.

But some large companies have shown the ability to remake themselves. Google, a consumer advertising business for decades, struggled for years to ramp up a cloud-server business but eventually got over the hump. By 2024 Google Cloud accounted for $43.2 billion in sales, more than 12% of Google-parent Alphabet’s total revenue.

Enterprise Traction

Apple has recently gained some traction with some enterprises in new areas. Its Vision Pro headset, for example, has struggled to gain traction among regular consumers—mostly due to its high $3,500 price tag—but has found a foothold in enterprise settings, such as helping airplane technicians repair engines.

During the cloud project’s early years, around 2020, its champion inside Apple was Abbott, a vice president responsible for a number of cloud projects, such as iCloud storage and mail. After his departure, his engineering org was placed under Jeff Robbin, a long-time engineering leader. Adrian Perica, Apple’s vice president of corporate development, who has taken over more of the company’s services business in recent years, would be responsible for running the cloud server business if it ever launched, the people said.

Without Abbott leading the charge, it isn’t clear if the project will proceed, though it still has backers inside the company.

The cloud service idea took root inside Apple with Project ACDC, which stands for Apple chips in data centers. As a first step, Apple last year launched Private Cloud Compute, which uses high-end Mac chips inside Apple data centers to run parts of its new AI system that can’t run on-device, such as more complex requests.

Siri was the first team to try out servers powered by Mac chips for text-to-speech capabilities. The servers provided a performance improvement in accuracy and cost reduction compared to traditional servers with Intel chips, said people involved in the project.

The ACDC team also recruited Photos and Apple Music to try out the system, said people familiar with the effort.

Internally, Apple has referred to its private cloud using its own chips as its own Amazon Web Services, the cloud computing division of Amazon, said people who worked on the project.