Apple may actually be winning the AI race

The big piece of Apple Intelligence was supposed to be a new version of Apple’s digital personal assistant, Siri, that works like the top AI chatbots from OpenAI and Alphabet. A smarter assistant is something Apple users have craved since Siri first arrived in 2011. But the project has been indefinitely delayed.

The new Siri is proving to be difficult because Apple came into the AI race by handicapping itself. It is the only Big Tech company that sees privacy and security as marketable features, not cost centers. Any implementation of the new Siri will need to meet Apple standards in this regard, and that is proving to be a big hurdle.

Apple’s strong preference is for all machine learning to happen on encrypted Apple devices, leveraging special units in Apple’s chips. Nothing is more private and secure. But the “frontier” language models that underlie ChatGPT and Gemini run in giant data centers, and are far too demanding for a phone. Much smaller models that can run on a phone don’t yet provide a consistently good enough user experience for Apple.

So we wait. Meanwhile, Wall Street seems to have moved on to a new narrative: It doesn’t matter if Apple is late to AI.

While most of Big Tech is sprinting to an AI future, Apple is running a marathon. Only time will tell who is right, but I share Apple’s long view of the AI boom. The company can take its time fitting AI into its products.

So far, hundreds of billions in capital expenditures are bringing Big Tech to the same place: AI models that struggle to stand out from each other.

It turns out that having the best AI models isn’t a moat, just a fleeting advantage. Many enterprise customers have said that AI language models are becoming commoditized, most recently Salesforce CEO, Marc Benioff.

“We use all of the large language models,” he said on the company’s Wednesday third-quarter earnings call. “They’re all very good at this point, so we can swap them in and out. The lowest cost one is the best one for us.”

There have been reports that Apple is in talks with Alphabet and start-up Anthropic to use their AI models, fine-tuned for Apple hardware, as a stopgap until the company can create its own high-performing models.

Apple is pacing itself, putting user experience and privacy above expediency. Amid everyone else’s AI battle, Apple created its Private Cloud Compute: open-source server software written in Apple’s programming language, running on Apple servers that sport Apple chips. As always, the company wants to own and control the whole stack, especially when it comes to privacy and security. AI chats can include very personal information, and Private Cloud Compute hides them from peering eyes, including Apple’s. At some point, an upgraded Siri will arrive and it will be more secure than any other chatbot.

Meanwhile, Apple is keeping its powder dry, increasing capital expenditures modestly to support Private Cloud Compute. By contrast, Meta Platforms, Oracle, Microsoft, and Google are polluting their once-pristine cash flow statements and balance sheets with hundreds of billions in combined capital expenditures for AI data centers. Meta stands out, spending around $70 billion on AI data centers this year, and promising more next year. It’s all for its own use, not to rent out in the cloud like the others. Debt levels are rising, and depreciation expenses from capex are beginning to mount. They will keep rising.

As Alphabet’s depreciation was up 41%, Microsoft’s 93%, and Meta’s 20%, Apple’s rose just 7% in the latest quarter. If a time comes where big capital outlays make sense, Apple has plenty of room to do that.

While Apple sorts out how AI fits into its software, the company’s strengths remain evident.

Wall Street analysts now agree that iPhone 17 will boost device sales growth to the highest level since fiscal year 2021. Services revenue continues to grow briskly, leveraging the more-than 2.3 billion Apple devices being used by customers. Because it isn’t raiding its cash flow statement like other big tech companies, the cash-return program will continue unabated. When Apple reports its first-quarter earnings, it will likely push all-time dividend payments and share buybacks past $1 trillion. Since 2012, the company has retired nearly half of its outstanding stock, raising per-share metrics by 79%.

And this whole discussion brings up a bigger question: How badly does Apple need AI features to sell devices? Since it became a mature category, people buy a new phone when they think they need a new phone. For better or worse, new features no longer drive smartphone sales.

During the Covid-19 lockdowns of fiscal 2021, Apple iPhone sales were up 39% from the year before, as customers needed new devices to work from home. Phone 16 was heavily marketed as the Apple Intelligence phone and sales were decent, but no one’s idea of a blockbuster. Now the iPhone 17 lineup is being sold in a more traditional Apple manner with focus on hardware, design, and camera—and it looks to be doing much better. Those fiscal year 2021 phones are five years old now in fiscal 2026, and people need a new one. It’s as simple as that.

Apple has plenty of time. Its investors should hold on for the ride.

Write to Adam Levine at adam.levine@barrons.com