At WWDC, Apple's AI strategy comes into question

One year ago, Apple announced Apple Intelligence, its response to the wave of sophisticated chatbots and systems kicked off by the arrival of ChatGPT and the age of generative AI.

Analysts said Apple's installed base of more than 1 billion iPhones, the data on its device and its custom-designed silicon chips were advantages that would help the company become an AI leader.

But it's been an underwhelming 12 months since then.

Apple Intelligence stumbled out of the gate while rivals like OpenAI, Google and Meta have continued to make headway launching new generative-AI models.

Now, investors are calling for Apple to do something major to catch up in AI, which is rapidly transforming the tech industry.

When CEO Tim Cook speaks at Apple's annual Worldwide Developers Conference in Cupertino, California, investors on Monday, fans and developers will want to hear how the company's approach to AI has changed. That's especially important after some Apple executives have said that the technology could be the reason the iPhone gets supplanted by the next-generation of computer hardware.

"You may not need an iPhone 10 years from now," Apple services chief Eddy Cue said in court last month in one of the government's antitrust case against Google, adding that AI was a "huge technological shift" that can upend incumbents like Apple.

The Apple Intelligence rollout was rocky. The first features launched in October — tools for rewriting text, a new Siri animation and improved voice, and a tool that generates slideshow movies out of user photos — were underwhelming. One key feature, which came out in December, summarized long stacks of text messages. But it was disabled for news and media apps after the BBC discovered that it twisted headlines to display factually incorrect information.

But the biggest stumble for Apple came in early March, when the company said that it was delaying "More personal Siri," a major improvement to the Siri voice assistant that would integrate it with iPhone apps so it could do things like find details from inside emails and make restaurant reservations.

Apple had been advertising the feature on television as a key reason to buy an iPhone 16, but after delaying the feature until the "coming year," it pulled the ads from broadcast and YouTube. The company now faces class-action suits from people who claim they were misled into buying a new iPhone.

Although Apple Intelligence had a rough first year, the company hasn't said much publicly. However, it's reportedly reorganized some of its AI teams.

JPMorgan Chase analyst Samik Chatterjee said in a note this week that investor expectations were set for a "lackluster" WWDC, as the company still needs to bring to market the features it announced last year, versus "addressing the more material issue of lagging behind other large technology companies in relation to advancements in AI."

Meanwhile, Apple is facing renewed competition in its core business.

OpenAI in May acquired the startup io for about $6.4 billion, bringing in former Apple chief designer Jony Ive to build AI hardware. The company hasn't provided details about its future devices.

Meta has made a splash with its Ray-Ban Meta Glasses, selling over 2 million units since launching in 2021. The devices use Meta's Llama large language model to answer spoken questions from the user.

And last month, Android maker Google said its Gemini models will become the default assistant on Android phones. The company showed Gemini doing things that go beyond Siri's capabilities, such as summarizing videos. Google also announced a $150 million partnership with Warby Parker to develop its own pair of AI-powered smart glasses.

A working Apple Intelligence is important for the company to encourage its users to buy new iPhones since devices released before the iPhone 15 Pro in 2023 don't support the suite of features. But AI hasn't been a key driver of sales for smartphones yet, and may not be for years, said Forrester analyst Thomas Husson.

"There's been some new cool features and services, but I don't think it has drastically changed the experience yet," Husson said.

Apple declined to comment.

For years, Apple didn't like the words "artificial intelligence." It preferred the more academic term "machine learning."

Apple focused its efforts on what could efficiently run on its battery-powered phones. The AI race, led by OpenAI and Google, was about bleeding-edge capabilities that required high-powered servers based on Nvidia graphics-processing units, or GPUs.

Then ChatGPT launched in late 2022, making AI the most important term in Silicon Valley. Soon after, Cook was telling investors that Apple was spending "a tremendous amount of time and effort" on the technology.

While Apple Intelligence is based on a series of language and diffusion models that the company trained itself, Apple hasn't publicly competed with Google, OpenAI, Anthropic, or other companies in what are called "frontier models," or the most capable AI systems that often have to be trained on large server clusters packed with Nvidia chips and fast memory.

The difference between the way Apple and its rivals approach AI can be seen in the company's approach to capital expenditures. Apple spent $9.5 billion on capital expenditures in its fiscal 2024, or about 2.4% of its total revenue.

The iPhone maker has rented the computing power needed to train its foundation models, it revealed last year, from Google Cloud and other providers. Apple's rivals are gobbling up billions of dollars of GPUs to push the technology forward.

Meanwhile, Meta, Amazon, Alphabet and Microsoft are planning to collectively spend more than $300 billion this year on capital expenditures, up from $230 billion last year. Amazon alone is aiming to spend $100 billion, and Microsoft has allocated $80 billion.

Apple's best chance to quickly catch up up may be to do what it's done many times in the past: Buy a company, and turn it into a killer feature.

It bought PA Semi in 2008 for $278 million, and turned it into the seed for its semiconductor division. Ahead of releasing the Vision Pro headset, Apple bought over 10 startups that worked on virtual and augmented reality. Even Siri was a startup before Apple bought it for more than $200 million in 2010.

With $133 billion in cash and marketable securities on hand as of the start of May, there isn't much Apple can't buy, assuming it could get regulatory clearance. However, OpenAI, Apple's current Siri partner, is likely out of reach with a valuation of $300 billion. And given OpenAI's new relationship with Ive to build hardware, there are reasons for Apple to slow the partnership down.

Anthropic, whose Claude chatbot is powered by one of the leading AI models, was valued at $61.5 billion in a funding round in March. In the Google antitrust case, Cue, a senior vice president at Apple, mentioned Anthropic as a potential replacement for Google as the default search option in the iPhone's Safari browser.

"They probably need to acquire Anthropic," said Deepwater Asset Management's Gene Munster, who has followed Apple for decades, in an interview.

That would be by far Apple's largest acquisition. To date, the most the company has paid is $3 billion, when it bought Beats Electronics in 2014 for $3 billion, part of an effort to catch Spotify in the music streaming market.

Apple could buy a company that's developing AI-based apps, even if they're on open-source or other company models. Perplexity, which is currently fundraising at a $14 billion valuation, has shown strong interest in the smartphone market and understanding of the value of being a default AI service.

In April, Perplexity announced a partnership with Motorola, and it's reportedly in talks with Samsung to integrate its technology into the South Korean company's version of Android, as well as take investment from the Apple rival. Cue mentioned that Apple had been in discussions with Perplexity about its technology at the May trial.

It's also possible for Apple to treat frontier AI like it treated search — as a service that can be filled with a partnership. Apple software chief Craig Federighi implied as much last year at a panel discussion during WWDC, saying that Apple would like to add other AI models, especially for specific purposes, into its Apple Intelligence framework.

Federighi specifically mentioned Google, whose Gemini can now fluidly speak to the user and handle input that comes from photos, videos, voice or text. Documents revealed during the Google trial showed executives from Apple, including Cue and M&A chief Adrian Perica, were involved in the negotiations over Gemini.

Apple has been designing its own chips since 2010, and with AI in mind since at least 2018.

The most powerful Apple M-series chips can tap into something called "unified memory," says WebAI co-founder David Stout, making them ideal for doing AI inference. Apple also includes good GPUs on its chips, he said. WebAI is building software that allows users to fine-tune, train and run big models on consumer hardware.

Stout's company has built clusters of consumer-grade Mac Studio computers to run big AI models, like Meta's Llama.

"We picked Apple Silicon because we think it's the best hardware for AI," said Stout, adding that in his company's tests, Apple's chips can output 100 million tokens per dollar spent versus 12 million tokens per dollar for an Nvidia H100.

Part of Apple's strategy for Siri, announced last summer, was to cajole its developers to build snippets of new code into their apps, which would make it simpler for Apple Intelligence and Siri to use the apps and get things done.

While Apple is still pushing "App Intents" — the same system powers stuff like lock screen widgets — the framework for how they work with Siri hasn't been released yet.

The threat that advanced AI like Google Gemini and OpenAI's ChatGPT represents to Apple was underscored by Cue at the trial last month. He suggested that the rise of AI threatened Apple's biggest business.

"AI is a new technology shift, and it's creating new opportunities for new entrants," Cue said at the trial last month.

There is a growing sense in Silicon Valley that sophisticated AI interfaces might one day replace smartphones and laptops with new devices that are designed from the ground up to take advantage of AI-based interfaces. That could mean people speaking or chatting with their devices to command AI agents, rather than tapping on touch screens or keyboards.

Upon joining OpenAI in May, Ive said he believes AI is enabling a new generation of hardware.

"I am absolutely certain that we are literally on the brink of a new generation of technology that can make us our better selves," Ive, the iPhone designer who retired from Apple in 2019, said in a video announcing that his company had been acquired.

Though AI represents a risk to Apple's current business, Deepwater Asset Management's Munster said the company has more time than many believe to adapt because of so many years of customer loyalty.

"This is still something that has existential risk to all these companies, including Apple, but I don't think we're at some break point in the next year around it," Munster said.