What was announced
Nvidia and Microsoft put Nvidia’s RTX Spark chip into Windows PCs at a Microsoft event in San Francisco on 7 October. Laptop preorders opened the same day, with availability on 16 October; compact desktops follow in November. Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI and Gigabyte are all building machines around it, according to Nvidia.
Microsoft’s own is the Surface Laptop Ultra. Nvidia says the configuration tops out at a 20-core Grace CPU connected to a Blackwell RTX GPU with up to 6,144 cores over a 600 GB/s link, with up to 128GB of unified memory and one petaflop of FP4 AI performance. It runs the full CUDA platform, which is the part that matters: the toolchain developers already use in the data centre now runs on the machine in front of them.
The prices
TechCrunch, reporting from the event, puts the Surface Laptop Ultra at two base models, one starting around $2,600 and a faster one around $3,700, with memory and storage options taking the price to $5,900. Microsoft said the highest-end configuration was already out of stock. The Surface RTX Spark Dev Box, a workstation, starts around $6,000 and ships with Visual Studio Code, GitHub Copilot CLI, WSL and PowerShell 7. Dell’s XPS 16 Creator Edition is $3,800, on preorder at Best Buy for delivery later in October.
Microsoft is offering up to $1,000 off for trading in a MacBook Pro, which tells you who the machine is aimed at.
The argument
“Just having a model doesn’t do much for anything,” Satya Nadella said at the event, per TechCrunch, adding that Microsoft is enabling orchestration “not just for our apps but apps for anyone’s agent.” Nvidia’s framing is the same one from the other direction: the point of a petaflop on a desk is that a model runs where the data already is, without a round trip to somebody’s cloud.
Nvidia also previewed a DGX Station for Windows, built on a GB300 Grace Blackwell Ultra Desktop Superchip with 748GB of coherent memory and up to 20 petaflops of FP4 compute. Nvidia says it can run models up to a trillion parameters locally. DGX Station previously ran only on Linux; Linux toolchains remain available through WSL. No release date was given.
What is actually new here
Local inference on a workstation is not new, and neither is a laptop with a discrete GPU. What is new is the memory. A 128GB unified pool is the difference between running a quantised model and running the one the lab shipped, and it is the specification that decides whether a developer keeps paying for API calls.
The gaming and creator claims come with the usual caveat. Nvidia’s figures — AAA titles above 100 fps at 1440p with DLSS 5, Reflex and G-SYNC — are the company’s own, as are Microsoft’s comparisons against a MacBook Pro, which Microsoft describes as preliminary internal testing.
What to watch
The compact desktops in November, and whether the Dev Box price holds once the first wave of preorders clears. Also whether the 128GB configuration stays scarce: Microsoft saying its top model is already out of stock on day one is either demand or supply, and the memory market this autumn makes the second explanation at least as likely as the first.
