At the Build conference, Microsoft unveiled its complete suite of Copilot+ PC solutions. Indeed, this is the correct way for AI PCs to be implemented:
Launching first on Qualcomm chips, because Apple has already proven that the ARM architecture has a significant advantage over x86 in the era of AI inference;
At the very least, Windows for ARM appears to have made significant progress. Microsoft is breaking free from the shackles of the Wintel alliance—something I noted in a previous article: Wintel cannot deliver AI PCs effectively;
Making AI an ever-present companion for users, rather than just a simple assistant concept. Emotional value is just as important as productivity;
However, I still maintain a clear-cut stance of cautious optimism regarding Microsoft's announcement:
Looking ahead two or three years, Microsoft could indeed equip a billion PCs with Copilot+, but how many users will actually use it? Just look at the current state of Office Copilot—which went from a stunning launch to being largely overlooked. This company has a history of "over-promising and under-delivering";
Obviously, these features are driven by the GPT-4o model. But why is it that all OpenAI members showcase MacBooks on social media? Why did OpenAI use MacBooks instead of Windows laptops during the GPT-4o launch event?
Yes, Windows accounts for nearly 80% of the desktop OS market share. However, if you look at the developer community, that share drops significantly—and developers are the bedrock of the AI PC user base;
OpenAI recently stated that in the past year, 3 million developers have driven the massive shifts in AI. Even if we optimistically assume that 3 million developers can drive ten times that amount in AI PC user demand, the potential user base is 30 million, not 1 billion;
The above assumption is not conservative;
Without a doubt, the next target for AI reconstruction is the Operating System (OS). Whether on desktop or mobile, in the context of multimodal applications, it's hard to interact with the outside world while carrying a laptop, but you can easily interact with the world via a smartphone or any new portable AI hardware;
As ARM-based SoCs become mainstream for desktop chips, the performance gap between desktop and mobile chips will no longer be significant. For productivity, the difference is smaller than screen size; for emotional value, mobile devices far outweigh desktops;
This is something Microsoft must do, but holding onto desktop OS market share in the AI era won't be easy, as the company spent over a decade failing to conquer the mobile market;
No company is truly safe because everyone needs to revolutionize themselves first: Google must reconstruct search, Microsoft must reconstruct Windows, and Apple must reconstruct all its hardware. Of course, they all have moats that provide enough time and space to adjust and respond;
Deploying models on the edge (on-device) might be the most suitable way for the broad consumer market. However, for PCs, the gap between the total PC user base and potential AI PC users might be at least an order of magnitude, whereas on mobile, this gap might be much smaller;
Of course, the prerequisite is that manufacturers deliver a truly excellent all-in-one solution.