What to expect at Microsoft’s Windows, Surface and AI event on October 7

It’s been two years since Microsoft launched Copilot+ AI PCs, its first major attempt to push AI-ready hardware, and the results have been mixed, to say the least. This month, the company is once again gearing up to embrace the virtues of agentic AI, the latest hot topic in the tech world, in San Francisco on October 7. Microsoft describes the event as “a conversation about how local AI will shape the next chapter of the PC,” and it will include notable guests like CEO Satya Nadella, head of Windows and devices Pavan Davuluri, and NVIDIA CEO Jensen Huang.

Given the PC world’s lack of enthusiasm for the Copilot+ initiative, the subsequent removal of intrusive Copilot+ features in Windows 11, and the lukewarm response to invasive features like Recall, Microsoft is approaching this event defensively. Can this really spark excitement for the latest hot trend in AI? Here’s what we expect to see in San Francisco next week.

NVIDIA is once again claiming it will redefine PCs with its RTX Spark, the company’s first major Arm-powered chip designed for massive AI models on Windows laptops and desktops. The Surface Laptop Ultra, announced in June, is one of the first RTX Spark-powered devices, and from the brief preview we got earlier this year, it appears to be a direct competitor to Apple’s powerful MacBook Pro. It sports a 15-inch mini-LED display with a peak brightness of 2,000 nits, a massive trackpad, and all the ports you could want.

But we still don’t know how much the Surface Laptop Ultra will actually cost, and there’s no concrete release date yet. I’d bet the company is saving these crucial details for next week’s event. I would also bet that it will be so expensive that only real AI diehards will want one. AMD recently launched its flagship AI desktop hardware, the Ryzen AI Halo, at $3,999, and NVIDIA’s existing Linux-powered DGX Spark systems are selling for $4,999 (down from $4,000 at launch).

Large AI models require a lot of RAM to run locally, which is one of the main reasons why these AI-driven systems are so exorbitant. To recap, the NVIDIA RTX Spark will feature 6,144 Blackwell GPU cores, 20 Arm CPU cores, and up to 128 GB of unified memory. The RTX Spark will also be less power hungry than existing AI hardware, consuming just 80 watts of power.

I don’t expect to be very surprised by the Surface Laptop Ultra. Aside from being one of the first systems to feature RTX Spark, this appears to be another high-end laptop. But I hope to find out what inspired this poetic interpretation of Microsoft Vice President Brett Ostrum’s notebook: “A machine like this should not stand still.” She should be pushed.

Microsoft isn’t exactly keeping this a secret: The company wants us to know that it will have some big news on agentic AI next week. This follows Meta’s launch of its Muse AI agent, which quickly reached 5 million downloads, but also arrived with a serious 0 day flaw. Just like with Recall, Microsoft could easily run into unforeseen privacy issues with a real AI agent, one that would potentially be able to see all your files, your browsing history, and act on your behalf.

We’ve already seen Microsoft’s concept of an AI agent platform in the form of Project Solara, which in demonstrations earlier this year was shown to work on a smart display and smart key badge. However, this still seems more science fiction than reality. The company also tried some weaker agent concepts in Windows, including Office Agent to help you create documents, as well as an agent to manage settings on Copilot+ PCs.

The central idea of ​​agentic AI – allowing an AI tool to work on tasks with minimal input – seems intriguing, especially to people who have already completely changed their workflows around modern AI models. But it’s also not something that will easily translate to people outside of the tech world. I can understand why so many programmers are excited about OpenClaw, but only a few people will want to manage an agent with an archaic command line interface.

Rather, I imagine Microsoft is trying to bring something like Perplexity’s Computer, which can automatically assign tasks to multiple AI models locally and online, right into Windows. The company recently revamped its Copilot app to make room for automated agentic workflows, and I wouldn’t be surprised if that continues to evolve with the rest of its agentic ambitions. And hopefully we’ll know more next week.