In June 2023, Apple chief executive Tim Cook walked onstage at the company’s flagship annual developers conference to announce, in the tradition set by Steve Jobs, “one more thing.”
Just six months earlier, OpenAI’s ChatGPT had launched, gaining a million users in just five days and triggering a new cycle in the tech industry. Generative AI, the technology promising a profound new level of machine intelligence, was not new. But the chatbot was the first product based on large language models to truly grab the popular imagination.
The proponents of generative AI say it portends a radical future where our devices can contextually understand vast amounts of information and offer dynamic, smarter, humanlike responses to our needs.
But the “thing” Cook went on to announce that June day was the Vision Pro, Apple’s mixed-reality headset which remains a niche product limited to the US market. As the dust settled, commentators were quick to pick up on something: not once, during a series of presentations from Apple’s top executives, had any of them even mentioned the term “generative AI”.
A year on, there are fears that Apple may have missed the boat on a generational shift. The likes of Google, Microsoft, Meta and Amazon have raced to capitalise on the technology, investing billions of dollars into the hardware required to power generative models.
Google’s launch of the Pixel 8 smartphone in October and Samsung’s Android-based S24 in January — both powered by Google’s Gemini family of AI models — have introduced a new concept into the industry lexicon: “the AI smartphone”. In the “AI PC” space, rivals like Microsoft, Qualcomm and AMD have staked a similar claim to early leadership.
Apple, however, is yet to launch an iPhone specifically marketed for the age of AI — although its new line of iPads powered by its M4 chip, launched in May, gave a hint at its ambitions.
The company has meanwhile had a run of bad headlines, as the EU and the Biden administration pursue it over alleged antitrust violations. Its shares have recovered from a slump at the start of the year, but after a major growth spurt as pandemic lockdowns were lifted, sales of the iPhone, Apple’s most lucrative product, are flattening. And in what risks being seen as a sign of the times, the company is vying with Nvidia, the darling of the generative AI boom, for the rank of the second most valuable US company.
When Cook emerges onstage at this year’s developers conference in Cupertino on Monday, he has the chance to change this narrative. The company is expected to unveil its new operating system, iOS 18, a major software update that will kick off its broader plan for generative AI.
But analysts say the stakes are high for Cook, who needs to position Apple as a genuine contender in a race set to define the next decade of technological growth.
“Apple needs to dispel the perception that it has been behind on generative AI,” says JPMorgan analyst Samik Chatterjee. “They need to be able to say: ‘OK, we have caught up with the rest of the industry.’”
When it comes to new technologies, Apple’s strategy has always been to perfect, rather than pioneer — refining existing ideas to offer the best user experience.
Take the iPod in 2001. Many MP3 players were already on the market when it launched, but Apple’s product was sleeker, smaller, and could hold hundreds more songs. Underpinning it was a more efficient, cheaper option for buying music, iTunes. Six years later Steve Jobs pulled off a similar coup with the launch of the iPhone.
Yet analysts say that with the advent of generative AI, the company is under unusual pressure to make its move sooner rather than later.
“With AI, it does feel as though Apple has had its hand forced a little bit in terms of the timing,” says Leo Gebbie, an analyst at CCS Insight, a tech research and advisory group. “For a long time Apple preferred not to even speak about ‘AI’ — it liked to speak instead about ‘machine learning.’”
Machine learning is a subset of AI that uses statistical analysis to find patterns in large data sets; the umbrella term “AI” includes several methods and techniques for helping a computer perform a cognitive function.
“That dynamic shifted maybe six months ago when Tim Cook started talking about ‘AI’ and reassuring investors. It was quite fascinating to see Apple, for once, dragged into a conversation that was not on its own terms,” Gebbie says.
On the company’s last earnings call, Cook said the company was feeling “very bullish about our opportunity in generative AI”, adding that the company has “advantages that will differentiate us in this new era”. Cook was seeking to shut down the perception that Apple is behind on generative AI.
The competitive race Apple is entering has three dimensions: developing chips that can power AI features more and more on its devices; creating so-called “killer” applications that will lure in consumers; and securing access to the most advanced generative AI models controlled by rivals Microsoft and Google.
The company is intensely secretive about its processes, but it has not been sitting on its hands. It has built out a team of top AI talent led by former Google Brain executive John Giannandrea, who was hired in 2018.
It already has a chip powerful enough to run an “AI smartphone” — at least, by the most commonly understood definition of such a product.
Exact specifications vary, but the term typically refers to a phone with a neural processing unit chip, or NPU, that is capable of running around 30 trillion operations per second, or Tops.
These NPUs with their neural engine make it easier to run some of the extremely compute-heavy AI applications locally on the device, rather than through the cloud. But they bring “huge technical challenges in terms of memory”, says Reece Hayden, an analyst at ABI Research, a technology intelligence company. “Even very small generative models have much higher memory requirements than any phone at the moment can sustain.”
Apple, which designs its own custom chips for its products, has had its own dedicated neural network architecture since 2017 and its latest A17 Pro chip on the iPhone 15 Pro and Pro Max exceeds the 30 Tops benchmark.
But the iPhone 15 Pro has only 8GB of RAM, compared to Samsung’s S24 smartphone, powered by Qualcomm’s chips, which has 12GB. Apple is not expected to launch a successor, the iPhone 16, until later this year.
This is a crucial battleground in the global AI race. Technology market research firm Counterpoint estimates that “AI smartphones” will make up 43 per cent of worldwide smartphone shipments by 2027, with 1 billion devices in use.
Google, Apple’s main rival in terms of phone operating systems with Android, is already forging ahead, rolling out tools for developers to make use of its Gemini models, powering new features like “circle to search”, and integrating its chatbot, Bard, into what the user does on the device.
“I firmly believe we are in a once in a generation moment to reimagine what phones are capable of with the advances in AI,” says Sameer Samat, president of Google’s Android ecosystem. “We are on a multiyear journey to rebuild Android with AI at the centre of it.”
The International Data Corporation is expecting 4.8 per cent year on year growth in 2024 for sales of generative AI Android smartphones, but just 0.7 per cent for Apple.
“If Apple wants to change this trajectory and accelerate this growth, it is crucial for them to outline a clear AI strategy in the coming weeks,” Nabila Popal, a research director at the IDC, says.
Apple’s challenge will be “convincing consumers why they need these new AI features and, more importantly, why it needs to be done ‘on their device’, to compel users to upgrade,” she adds.
One concrete step Apple can take to show it is serious about the potential of generative AI is by using it to enhance Siri, the digital voice assistant it introduced in 2011.
The initial wave of AI chatbots led by ChatGPT, while catching the eye of consumers, is only one application of the technology — with the input and output both being text. An enhanced Siri might be attuned to user preferences and capable of carrying out a specific set of tasks very well, such as sorting through emails.
Within the limits of its existing hardware, Apple can use personal data kept locally on the device to craft these kinds of personalised experiences for its users, says Tim Bates, a professor at the University of Michigan-Flint College of Innovation & Technology. “I talk about this as ‘narrow AI’,” he says.
This so-called “on-device” premise has the added benefit of protecting user safety and privacy, as consumers are unlikely to want AI applications training themselves on their personal information and exporting it to the cloud. Running features locally also cuts out the lag involved in generating responses from a remote server.
“Siri is really the perfect ‘flavour’ of interactive AI,” Bates adds. “An individual can control their data, talk to the AI, and get things done, and not be afraid it’s going to be sucked out of the device.”
The killer use case of the AI iPhone, however, will come when Apple can offer a “conversational assistant that is fully integrated with everything on the device and then interfacing with some kind of expert agent in the cloud,” says Wamsi Mohan at Bank of America.
This is where the expected partnerships with OpenAI — and possibly Google — come in.
While Apple has been working on its own generative models, it cannot rival the most advanced models based on trillions of parameters of data, so it needs to strike deals.
“Apple is going at it in two different contexts. One is on the iOS devices, which nobody is going to do outside of Apple,” says JPMorgan’s Chatterjee. “The cloud aspect will involve separate partnerships — Apple doesn’t want to go and invest in data centres running heavy AI models.”
This would allow Apple to introduce generative AI to the rest of its software ecosystem. Dylan Patel, chief analyst at research group SemiAnalysis, says he expects Apple will have three layers to its AI offering, making use of multiple models: the smallest running locally on its devices, “medium sized” models running on their own servers, and “large models” with service providers like OpenAI and Google.
Apple’s built-in advantage is its role as a hardware “middle man” between the hundreds of millions of iOS users around the world and the developers using its software ecosystem. It is able not just to curate how consumers experience new generative AI features, but also how developers build them, ABI’s Hayden says.
Developers are expecting new tools from Apple, says Adam Smart at mobile software analytics firm AppsFlyer, both in terms of specific software development kits for building out AI features, and new application programming interfaces that allow them to tap into Apple’s own features like Siri.
“An AI partnership has potential to change the game when it comes to streamlining common, everyday tasks people use their Apple devices for, such as crafting and editing emails and text messages, or simplifying voice-to-text search functions in travel and shopping apps,” he says.
The developers conference ought to shed light on this too, says BofE’s Mohan. “I think Apple will provide some of the hooks that are necessary to enhance productivity with AI.”
These moves might allow Apple to draw level with some of its rivals, but not yet surpass them. At the moment, it is neither pioneer nor perfectionist.
To the extent that progress rests on partnerships with large partners, the precise contours of any deal between Apple and OpenAI are unclear. Whether the two companies make a cultural fit, and how long the marriage can last, is an open question. The potential pairing is all the more striking given Cook’s decision not to even acknowledge the company just a year ago.
But experts caution it would be premature to count Apple out. The AI revolution is still in its infancy, and none of the Big Tech companies have yet managed to graft generative AI on to a killer hardware product.
“To use a sports analogy, I think we are in the first minute of the first half,” says Jason Banta, a corporate vice-president at chipmaker AMD. “It’s still very early.”
Read the full article here