According to Tim Cook, these are the first smartphones “designed for artificial intelligence”. During the Keynote September 9, 2024, Apple’s big party where the firm presents its new iPhones, the boss emphasized AI. Here, the acronym does not mean Artificial Intelligencebut Apple IntelligenceThe technology, which will be featured across the entire iPhone 16 range (the 16, 16 Plus, 16 Pro and 16 Pro Max) is expected to revolutionize the use of Apple phones through numerous features.
The release of the iPhone 16 finally positions the company in the AI race. Above all, it reassured the tech world, which had been surprised for months by the company’s silence on the technology, and feared that it had fallen far behind in the field. A shame for this pioneer of intelligent assistants who made his mark with Siri. However, Apple has an atypical strategy for entering the AI race.
Apple’s “three-blade razor”
The first pillar of its policy: an AI model developed by the companywhich will allow certain tasks to be performed “locally”, directly on the devices. A tool designed for the most basic user requests – rewriting messages, light photo modifications – and which will enrich the capabilities of Siri.
For more complicated queries, you will have to go through a second layer: a cloud service specially developed by Apple. The advantage of this system is that user data is thus protected by Apple, it is not communicated to third parties. This cloud, called Private Cloud Compute, will run on powerful Apple silicon processors. Top floor of the building: the company delegates the heaviest tasks to OpenAI. Apple has signed an agreement to integrate ChatGPT into iPhones. Users can thus launch queries directly from their phones.
Apple has redoubled its efforts in AI. In February 2024, the firm announced that it would close the Apple Car division, its electric car project, in order to reassign 2,000 employees on the subject. The company also poached a dozen artificial intelligence experts from Google in April to strengthen its teams.
Its three-layer strategy, however, shows that it does not yet have a sophisticated LLM (large language model). “If they had a big model capable of competing with ChatGPT or Gemini, they would have put it on their private cloud,” observes Hanan Ouazan, partner at Artefact in charge of AI offerings. Focusing on “mini-AIs” presents a risk: that the gap widens with more sophisticated models and that “Apple’s small models become outdated,” points out Pascal Malotti, head of strategy for the Valtech consultancy.
But this cautious approach has its merits. Because with small artificial intelligence models, Apple can already meet many everyday demands. The OpenAI partnership allows Apple customers to be offered the best artificial intelligence on the market, without exorbitant investment. AI training, which requires a lot of computing power, is in fact excessively expensive. Around $100 million for training GPT-4. And the bar will be set even higher with new generations of AI: studies put it at around a billion dollars in a few years.
“The big names in AI are way ahead. And in this field, the competition is moving very quickly,” Hanan Ouazan also points out. Even Google, which has been in the sector for many years, is struggling to catch up with OpenAI.
Apple is not competing with OpenAI
The risk of a financial bubble still hangs over the industry. “Apple does not want to take the risk of investing in expensive infrastructure that may not have a return on investment,” Pascal Malotti emphasizes. Relying on the technologies of other existing players remains the most economical solution.
Access to training data would finally be a dilemma for Apple. Not that the firm lacks theoretical means to collect it: its products are in the pockets of two billion people. But the firm has for years made it a point of honor to preserve the privacy of its users at the heart of its discourse. However, “LLM training is a disaster for data protection”, analyzes Pascal Malotti.
So many reasons why Apple is not panicking about the AI race. “Apple is first and foremost a hardware seller, more than software. For them, AI is not that important,” emphasizes Hanan Ouazan. The company could well continue to delegate certain tasks to AI partners while waiting for the bubble to burst and the market to consolidate. “LLMs are like the A380,” points out this partner at Artefact. “It’s impressive, but not all airlines need them.”