Key Highlights
- The Surface Laptop Ultra starts at USD 2,599, is open for preorder and begins shipping on Oct. 16.
- The laptop holds up to 128 gigabytes of unified memory and delivers up to one petaflop of computing power.
- Nvidia raised the price of its DGX Spark AI desktop by about 75% to USD 6,950 amid rising memory costs.
- Microsoft is not among the top six PC vendors by unit shipments, according to a research firm estimate.
Price, Specifications and Timing
Microsoft (NASDAQ:MSFT) said its new laptop built around an Nvidia AI chip starts at USD 2,599, as the software maker tries to bring more artificial intelligence capability to consumer devices. The Surface Laptop Ultra is available for preorder and begins shipping on Oct. 16, a Microsoft executive said at a company event in San Francisco on Wednesday. Microsoft first discussed the machine in May and has now provided specifics.
The laptop contains Nvidia's RTX Spark chip and carries up to 128 gigabytes of unified memory, with up to one petaflop of computing power. Nvidia's chief executive said at the event that this matches the compute of the company's DGX-1 Data Center server from 2016. A magnetic USB-C port echoes the charging connector on Apple's MagSafe line.
The pitch centers on running models locally. Users can install and run a range of models on the device without worrying about monthly cloud bills. At the event, a presenter showed how apps can route some AI coding requests to the computer's local GPU, signaled by a green leaf icon, while sending others to the cloud.
Strategy in a Crowded PC Market
Windows is the most widely used PC operating system, yet Microsoft ranks outside the top six PC vendors by unit shipments, a research firm estimates. Microsoft has sold its own computers since 2012, and its chief executive said the company's position is evolving, pointing to a "frontier ecosystem taking maybe slightly different contours two years from now, five years from now, 10 years from now." He added that the classic idea that distributed computing stays distributed holds more strongly now.
Microsoft is not alone in moving AI workloads toward the device. Apple and Google are also pursuing ways to let users run more of those workloads locally rather than in the cloud. Microsoft previously tried to use the AI cycle to lift device sales, introducing the Copilot+ PC standard in 2024, which required neural processing units that meet a minimum AI performance bar. A company spokesperson said it has since simplified product naming to lead with the Surface Brand while continuing to offer premium hardware and hybrid AI experiences.
Microsoft's own AI models are less widely used than leading offerings from OpenAI and Anthropic and do not lead on at least one performance benchmark, and a recent survey found developers have used the alternatives more. The company is also updating the Windows 11 setup experience to help people get started with the open-source OpenClaw AI agent software.
The Memory Cost Question
Analysts said price is the largest open question after a memory-chip squeeze lifted costs for desktops and laptops. When Microsoft first pitched on-device AI two years ago to save costs, laptops were mostly priced under USD 2,000. In recent weeks Nvidia raised the price of its DGX Spark AI desktop by about 75%, to USD 6,950, because of rising costs for its 128 gigabytes of memory.
One analyst at a research firm said the software was not ready two years ago but the hardware was, and that now the software is ready while the hardware is too expensive to run locally for many users. "It's becoming this thing where only the people who have the budget can really afford to run AI locally," the analyst said. Another observer wrote on Social Media that the announcement might get him back to Windows.
Markers Around the Launch
The details to follow are the Oct. 16 shipping date, the final pricing of configurations with 128 gigabytes of memory, the trend in memory-chip costs across PCs, and how developers use the split between local GPU processing and cloud routing.






Please wait processing your request...