The Surface Laptop Ultra is an AI monster with a laptop problem


There are further tradeoffs. Users will be able to run models locally, but the Surface Laptop Ultra ships with a 140W power adapter. That’s a signal that it will consume much more power than typical laptops, though not gaming laptops. (That’s another question: is this a gaming laptop? Microsoft showed just a single Gears of War game running on the Ultra, with no indication of frame rate.)

Nevertheless, I left the Windows and Surface Event with a question that no one at Microsoft could answer: What will be the battery life of these laptops? In one demonstration, Microsoft employee Kayla Cinnamon proudly put the Surface Laptop Ultra into “airplane mode,” showing how the laptop could run entirely disconnected from a network.

(Microsoft’s official stance: “Battery life details for Surface Laptop Ultra will be shared closer to availability,” reads the Ultra’s spec page. “It varies based on usage, settings, configuration, and other factors.”)

But these laptops may not even work on airplanes. For one, 140W is more than some airplane power outlets can supply—110W to 150W is apparently typical. (Edit: Jason Rabinowitz, cohost of the AVTalk podcast and an avid airline enthusiast, says that 60 to 100 watts is actually more common.) Will you be able to run AI agents continuously throughout a 13-hour flight from San Francisco to Taiwan? I very much doubt it. We don’t even have a reliable way to test the battery life while churning tokens.

All of these unknowns left the analysts I spoke to shaking their heads. Who’s going to buy these devices? The $2,599 price is the base, which means as users scale up to more and more memory—to run larger and more powerful AI models—the price will balloon with it. We’re still in 2026, a time when the $699 Apple MacBook Neo blew up headlines.

“I think it’s very obvious that no one wants to spend all this crazy money [on cloud AI],” said analyst Bob O’Donnell, principal at Technalysis. But how users want to divvy up their work between slower local AI and faster, more expensive cloud models remains to be seen.

Exit mobile version