Apple’s new AI is made in Google data centers

For years, Apple has leaned on Google and Amazon’s cloud services to store data for its products. For example, when Apple device users do iCloud backups they are often stored in Google’s data centers. It’s not a deal that either company tends to talk about, but it gets Apple access to thousands of Google’s machines, helping it provide many of the software features iPhone users love and rely on.

TPUs OMG

When it came to training the AI models that power the iPhone maker’s new Apple Intelligence, the company asked for additional access to Google’s Tensor Processing Units for training. TPUs are chips specifically designed for AI and Google rents these out via its cloud service, as an alternative to Nvidia GPUs.

Apple’s request created a scramble inside Google in April, when Googlers became aware of technical issues that could have stop them delivering what Apple wanted on time.

It was known internally as an “OMG,” a Google term for one-off urgent incidents that don’t quite warrant a code red. A war room was convened inside Google, according to a person with direct knowledge of the incident.

“Bigfoot”

The team delivered for Apple after a few very long days, the person said. But it was a close call that could have spelled bad news for Apple, which has earned the nickname “Bigfoot” among Google Cloud employees because of how much of Google’s data centers it uses.

Spokespeople for Google and Apple did not respond to a request for comment.

The partnership underscores how far behind Apple still is in the generative AI race. Most of the impressive features of AI models must be at least partly handled in massive energy-sucking data centers, which companies such as Microsoft, Google and Amazon have spent years building. As a result, Apple is having to lean more heavily on these rivals as it tiptoes into the AI race.

As some eagle-eyed Apple watchers spotted, Apple’s technical documentation hints at the Google partnership, with a mention that its AI models were trained using a combination of methods “including TPUs and both cloud and on-premise GPUs.”

When users actually try out Apple’s new AI features, much of the work will happen on the device itself. More intensive tasks will be handed off to what Apple is saying are special data centers running new Apple-designed silicon. Where these servers are actually located remains unclear.

Unexpected partnerships

But Apple is faced with a new reality where cloud computing and the chips to train AI models are a hot commodity, forcing it to tap partners that it also competes against.

For example, Apple’s deal with OpenAI will give users access to a more advanced chatbot than Apple can offer in the form of ChatGPT. It’s also a boon for OpenAI, which gets new access to Apple’s vast user base.

The AI wars are forcing tech companies to forge these crucial, and sometimes unexpected, relationships. On Tuesday, Microsoft and Oracle announced a new deal that will give Microsoft access to Oracle’s cloud servers for handling some OpenAI workloads. Previously, OpenAI ran exclusively on Microsoft’s servers.

Bloomberg previously reported that Google and Apple were also in talks to bring Google’s Gemini AI to iOS devices. So far a deal has failed to materialize, but that doesn’t mean it won’t happen.

Speaking in an interview after the keynote, Apple executive Craig Federighi made it clear that Apple is open for business when it comes to AI partners, and even namechecked Google.

“We want to enable users ultimately to choose the models they want, maybe Google Gemini in the future,” he said. “Nothing to announce right now.”