Microsoft’s pitch at its October 7 Windows event was that the next wave of AI will run on your own computer, not just in the cloud. The headline product was the Surface Laptop Ultra, built on Nvidia’s new RTX Spark chip with up to 128GB of memory, starting at $2,599 and shipping October 16, Engadget reported. Microsoft also showed Copilot acting on local files and running AI models on the device, and announced a $5,999 Surface RTX Spark Dev Box for heavier AI work.
Marques Brownlee summed up the launch in one line:
New Surface Laptop Ultra, powered by RTX Spark starts at $2599 https://t.co/GFLe30Za71
— Marques Brownlee (@MKBHD) October 7, 2026
The starting price is not the number that tells you what local AI costs. The memory upgrade is.
Local AI is a memory problem
Running a large AI model means holding it in memory. The bigger the model, the more memory it needs, which is why Microsoft’s top configuration offers 128GB of unified memory. According to ServeTheHome’s breakdown of the lineup, the base model with 24GB of memory costs about $2,600. A 20-core version with 32GB and a 1TB drive costs about $3,700. The same 20-core chip with 128GB and the same 1TB drive costs about $5,900.
That is roughly $2,200 for the extra memory alone. ServeTheHome noted that some of Microsoft’s own models are recommended for systems with 128GB or more, and described RAM’s current cost as extreme.
That cost has a familiar cause. Memory prices have soared as AI data centers buy up supply, the same squeeze we covered in our piece on phone prices and that pushed Apple to raise the price of every iPhone by at least $100. Local AI asks the buyer to pay for one of the fastest-rising costs in electronics, and to buy a lot of it.
Who carries the cost
There is an economic shift hiding in the local AI pitch. When you use an AI model in the cloud, the company running it pays for the chips, the memory and the electricity, and recovers that cost through subscriptions or bundled services. When the model runs on your laptop, much of that cost moves to the hardware you bought. Microsoft promoted a hybrid approach in which Copilot runs some models locally and taps the cloud for extra compute. Every task that stays on the device is compute the cloud provider does not have to supply.
That is not necessarily a bad deal for users. It is a different one, and worth seeing clearly. A $5,900 laptop is a large up-front purchase of the capacity a cloud service would otherwise rent to you over time.
The case for running AI locally
There are real reasons to want AI on your own machine. Data that never leaves the device is easier to keep private. Local models work offline and can respond faster without a round trip to a data center. For developers, a machine that can run large models without paying per request can save money over time. And the Surface Laptop Ultra is not only an AI box: Engadget noted gaming support including ray tracing and DLSS, and Microsoft highlighted a replaceable SSD.
On price, ServeTheHome called the top configuration surprisingly competitive, noting that the cheapest 14-inch MacBook Pro at the same memory level costs $800 more. High-memory machines are expensive everywhere right now.
There is also a significant shift in who makes the chip. The RTX Spark is an Arm-based Nvidia processor, according to ServeTheHome. Microsoft’s most powerful Surface does not use Intel, AMD or Qualcomm, which says a lot about where the PC industry thinks AI performance will come from.
What to watch
The question for the next year is not whether AI can run on a laptop. It clearly can. It is whether enough people will pay thousands of dollars extra in memory to do it, while similar capabilities are available from cloud services for a monthly fee. If local AI catches on, expect memory to become the most important spec on a PC’s box. If it doesn’t, the 128GB laptop will remain a developer’s tool, and most people’s AI will keep running on someone else’s hardware.
Related reading: Memory Prices Are on Track to Rise 400%. Your Next Phone Is Paying for AI. · Google’s $899 Googlebook Isn’t a Chromebook Successor.
