After the Build conference, I wrote that Microsoft had sped past Apple with the announcement of its Copilot+ PC, which provided a fundamentally new way to use artificial intelligence in the consumer experience.
However, I also wrote that the race was not over and we would have to wait until WWDC 2024 to see how Apple responds. And with the announcement of Apple Intelligence, Apple proved that it might not be at the forefront of generative AI research, but it sure has a few important cards to play.
Both companies (along with Google) are locked in a race to dominate the market for the consumer AI experience. And now that we have a preview of their strategies, it’s time to update the strengths and weaknesses of each.
Apple’s historic strength is its vertical integration and holistic experience. With full control over the entire ecosystem of Apple products, it can deliver AI experiences that seamlessly transfer from your iPhone to your Mac, iPad (and possibly Apple Watch in the future). In contrast, the Copilot+ PC is detached from your phone, unless Microsoft finds a partner.
The features of Apple Intelligence were not crazily superior to those of Copilot+ PC. You get the same kind of tools, including on-device language models that can perform tasks such as write emails, summarize text, and search the content of your images, and diffusion models that can generate images and emojis. But the company has done a great job of getting the most out of on-device hardware with the right techniques and algorithms.
Apple is also expanding its cloud infrastructure with Private Cloud Compute (PCC), which sends AI workloads to the cloud when the device can’t handle them. And when PCC can’t handle a task, it asks for your permission to send it to OpenAI’s GPT-4O, free of charge.
This turn to the cloud marks a shift from Apple’s privacy-focused product development. Apple has always taken pride in not collecting user information. And in WWDC, the Apple execs went into detail about how PCC was designed from the ground up with privacy in mind. However, in my opinion, once the data leaves your device, you have already crossed a line, regardless of how/if you store and process it. After all, Microsoft and Amazon also have very secure cloud offerings with Department of Defense approvals.
The OpenAI partnership is even more questionable, since Apple is providing it for free to its users and is reportedly also not paying anything to OpenAI. There must be some incentive for OpenAI besides just distribution to Apple’s user base.
PCC and Apple’s partnership with OpenAI also marks another important point: We still have a long way to go before we have fully capable and reliable on-device generative AI.
On a separate note, Apple has now created another incentive for its customer base to upgrade their devices on the back of a slump in iPhone sales. Apple Intelligence only works on iPhones with A17 processors and MacBooks and iPads with M-family chips. This means that if you don’t have the iPhone 15 Pro or Pro Max, you won’t be able to use Apple Intelligence on your phone.
However, in comparison to Microsoft, Apple still has the advantage on the device side. As I wrote in my previous article, “Apple’s strategy has always been to create a very polished product with cutting-edge technology and lots of headroom for future applications and software features.” This means that there are already millions of MacBooks and iPhones that can run Apple Intelligence. In contrast, Microsoft has set 40 TOPS as the minimum requirement for a Copilot+ PC, which means fewer people will be able to experience the on-device AI without purchasing new PCs.
On the other hand, Microsoft has the advantage of superior AI capabilities. Microsoft’s scale in training and serving generative models is unmatched. It has exclusive access to OpenAI’s models, and is also hosting the ChatGPT models that Apple will be serving to its users. Something tells me that Microsoft will be the big winner of Apple’s partnership with OpenAI.
While Apple continues to eke out every ounce of possible performance out of its on-device AI, it will still have to deal with the reality of today’s models: Superior models will require both better data and better compute infrastructure. Apple has neither, unless it starts paying for more licensed data, mines its users’ data, spends more on its cloud infrastructure, or goes deeper into partnerships with the likes of OpenAI.
Bottom line for round two: Apple has made a good comeback, underlined by the sudden jump in its stock price. However, everything so far is speculation. Next round will be when users get to try and compare both technologies.