Battle of AI Hardware Decided

Apple, which has hitherto remained reticent about its AI strategy, finally laid out its homegrown AI plan at the recent WWDC.

The scheme proposed by Apple is indeed commendable, perfectly leveraging the advantages of its ecosystem. By wielding user privacy as a weapon, Apple has set the tone for its future development. Reflecting overnight, let's begin with Nvidia, the concern that most people have.

"Greater Effort Required from Nvidia"

Apple's AI plan clearly states: "I am not a customer of Nvidia's high-end servers, I can serve my users with my own chips!" Looking at the AI model architecture that Apple has revealed, it is clear that Apple is engaged in subtraction. By optimising the model through fine-tuning, pruning, precision reduction, and focusing on specific usages, the aim is to fit it into Apple's own equipment, indirectly severing dependence on servers. This is undoubtedly bad news for Nvidia. Apple's users account for 10% of PCs, 20% of mobiles, and 38% of tablets, meaning a lost opportunity to directly serve these high-end users. Nvidia is building a business model where Token equals Money, but Apple's current strategy does not endorse this story.

Apple believes AI is merely a means to enhance user experience, and the cost should be borne by the service provider. Of course, this is because Apple has pricing power over its own products, so this scheme is just a way for them to shift costs onto consumers. However, I believe Nvidia still has a chance to convince Apple. Focusing on the future, multimodal AI models will undoubtedly be thousands of times better than small models, so the Token equals Money story still has a chance to be told in Apple's Keynote. But at the current stage for Nvidia, Apple has become a competitor vying for TSMC's high-end process. It can be predicted that Apple will need a large number of M-series chips to expand its private cloud, and it can also be predicted that the iPhone16 will experience explosive growth at the end of the year. Both will require a huge amount of high-end chips.

On the one hand, knowing that they are losing Apple as a customer, they need to look for more possibilities. On the other hand, they must compete with Apple to secure more capacity.

"A Bright Outlook for AI PC in Enterprise market"

The WWDC conference still focuses on personal users that Apple is most concerned about. The announced features and updates all emphasise how to improve personal efficiency. This is undoubtedly excellent news for the Microsoft AI PC camp, which targets corporate users. Without Apple competing for corporate users, Microsoft only needs to keep an eye on Google's movements. Moreover, the comprehensive update of Apple Intelligence is a bit late, and this autumn will only support English. Other languages are expected to be available next year. This golden period of no competition will provide the AI PC camp with an excellent sales window.

Enterprises are very willing to pay for work efficiency, especially since many enterprises are already customers of Microsoft's full suite of products. A machine integrated with hardware, software, and AI will be the most compelling reason for everyone to pay. However, there is only one major concern. That is the lack of software support for the arm base chip architecture. Although the trend of software services is to deploy on browsers, many heavyweight enterprise-level software is still deeply bound to the X86 architecture platform (SAP, PowerBI). Microsoft still needs to give a ladder to accelerate these software service providers to move to the arm platform. Like the compiler that Apple launched when it switched to the M series chips, allowing these software to run smoothly in the new environment.

Next, let's talk about my thoughts on Apple's AI plan this time.

"Apple's Command of User Privacy as an Imperative"

Apple, unsurprisingly, remains the company we've always known. Apple's strength has always been its ability to focus on key areas and establish benchmarks. From its early fondness for custom hardware specifications to its recent establishment of what could be the most crucial privacy usage regulation in software industry history. Why do I make such an assertion? There are several points to consider.

"Regulation to Unfetter Privacy Protection"

In the past, the interaction of information between applications was restricted by the system's privacy protection policy. Each application had to have its own account system to verify the user's status and the scope of their usage. This meant that every service was an independent entity, each needing to maintain its own user data and authorisation. However, now, through Apple's intents API, the authorisation of services can be consolidated at the Siri level, with Siri handling related privacy authorisation issues. In essence, Apple has opened the door for all applications to communicate with each other. However, the prerequisite is that your application service must comply with this privacy regulation. Furthermore, the exchange must be mediated by Siri.

Now, Apple has stepped forward and declared, "I am now allowing you to interact with each other, but it must be under my supervision." In the future, you could tell Siri, "Call a car for my mother, she's almost at the high-speed rail station." Siri can then autonomously review your address book, find your mother's name and phone number, input it into the Uber call car message, and complete the call through Uber, without even needing to open the APP. I believe e-commerce applications will be even more interesting. Thus, if your application relies on traffic or advertising revenue, you may need to consider other business models, as some of the demands may be fulfilled at the Siri level.

"Private Cloud Computation"

Another interesting architecture unveiled at the WWDC is the new private cloud computation. Apple has established a new private cloud computing system. Although this plan was originally intended to supplement local AI computation deficiencies, Apple is very cautious in designing privacy protection architecture for data processing on the cloud. To win public trust, Apple unusually allows some authorized developers to transparently view the internal information of cloud computation. This not only shows that we are transparent and unafraid of others looking, but also reveals Apple's full confidence in its privacy design for cloud computation. This leads me to envisage a possibility.

I have long been puzzled by Apple's lack of cloud services (not the consumer-oriented iCloud, but those for business use). Given the number of Apple developers, maintaining a well-functioning, smooth cloud service should not be difficult (perhaps the profit margin is too low?). However, with this opportunity of private cloud computation, could it be possible that in the future it would be open to other applications, allowing other services to be deployed in Apple's private cloud? Or is the issue the opposite: in the future, if you don't put your service on Apple's cloud computation, you won't have access to Apple's users?  It is for sure the innovation on Apple App will be the next battlefield.