Honest fence (read first): This brief is built on NVIDIA’s blog and two Microsoft blogs, all dated October 7, 2026, the day of Microsoft’s Windows AI and Surface event in San Francisco, where Jensen Huang and Satya Nadella spoke. RTX Spark itself is not new. NVIDIA unveiled the chip in June at GTC Taipei. What’s new is preorders, prices and ship dates, plus Microsoft’s agent-security layer for Windows going generally available. Prices are U.S. MSRP. The posts give no Canadian price. Every performance line (“a petaflop,” “models exceeding 120B parameters,” “matches the intelligence of many cloud models”) is the companies’ claim. No independent review exists yet. The embed is Windows’ official event replay (ilmBGeGldrI; RTX Spark laptops at 1:00:00, Surface Laptop Ultra at 50:20).

Windows — “Something new is coming from Windows | October 7” (event replay, published October 7, 2026). This is Microsoft’s own event video, not an independent review; prices and ship dates cite the NVIDIA and Microsoft posts below, and the local-AI speed claims are untested.

Editorial Thesis

For two years, “run the model yourself” meant a Linux box with a big GPU, or a Mac with lots of memory. RTX Spark is NVIDIA and Microsoft’s bid to make it a Windows laptop you can order today. The pitch is memory, not just speed. Up to 128 GB shared between the CPU and GPU is what lets a 100-billion-plus-parameter model fit on a laptop at all.

For practical readers, the question isn’t whether the chip is fast. It’s whether a $2,599-and-up laptop replaces a cloud AI subscription for the work you do, and whether your Windows apps run well on an Arm-based machine. Neither can be answered from a launch blog.

Honest Verdict

Believe the prices, dates and specs. They come from Microsoft and NVIDIA directly: Surface Laptop Ultra from $2,599, available October 16; Surface RTX Spark Dev Box $5,999, U.S. only, shipping in November; up to 6,144 GPU cores, a 20-core Grace CPU and 128 GB of unified memory. Believe that MXC is generally available. Microsoft says so. Treat the AI performance claims as marketing until reviewers test shipping units. Note that the 128 GB configuration is not the $2,599 one: retailers list the top Surface Laptop Ultra at about $5,899. DGX Station for Windows is a preview with no price or date.

What the primaries actually say

NVIDIA (Oct 7)

  • RTX Spark “laptop preorders open today and available October 16. Compact desktops will be available for sale in November.”
  • Hardware: an NVIDIA Blackwell RTX GPU with up to 6,144 cores and an up to 20-core Grace CPU connected at 600 GB/s. “One petaflop of FP4 AI performance and up to 128GB unified memory.”
  • Claim: runs “Qwen 3.8 Flash Next, a 125B model … that matches the intelligence of many cloud models, unmetered and without sending data to the cloud.”
  • System makers: Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI and Gigabyte.
  • DGX Station for Windows (preview): GB300 Grace Blackwell Ultra, 748 GB coherent memory, up to 20 petaflops FP4, “enough to run models up to a trillion-parameter scale locally.” Sign-up only.

Microsoft (Oct 7)

  • Surface Laptop Ultra: “starting at $2,599 (MSRP) … with availability beginning October 16.” Microsoft says it can “run AI models exceeding 120B parameters locally, with up to 1 petaflop of AI performance.” 15-inch display, under 18 mm thick and under 4.5 lb.
  • Surface RTX Spark Dev Box: “$5,999 (MSRP), exclusively on Microsoft.com in the U.S. and will begin shipping to customers this November.”
  • Trade-in: up to $1,000 back for an eligible MacBook Pro, through Microsoft Store online in the U.S. and Canada.
  • Windows: “Microsoft Execution Containers (MXC), now generally available for Windows,” tied into Microsoft Agent 365. Microsoft describes it as OS-level containment so background AI agents can be secured, observed and governed. Local models named for RTX Spark include DeepSeek V4 Flash (284B) and an upcoming NVIDIA Nemotron model with more than 70B parameters, quantized to 2-bit to fit in just over 20 GB. Windows ML adds llama.cpp support.

Claim vs proof scoreboard

ClaimProof today
RTX Spark laptops can be preordered nowNVIDIA + Microsoft blogs (Oct 7)
Surface Laptop Ultra starts at $2,599, ships Oct 16Microsoft Devices Blog (U.S. MSRP)
128 GB unified memoryTop configuration only (~$5,899 at retail, secondary)
“Runs 120B+ models locally” / “a petaflop”Company claims. No independent benchmark yet
“Matches the intelligence of many cloud models”NVIDIA’s claim about one model
MXC is generally availableMicrosoft Windows blog
DGX Station for Windows is comingPreview. No price, no date
Canadian priceNot published in these posts

Who should care

  • Developers and small AI teams paying cloud GPU bills for experiments. Wait for the first reviews (from October 16), then compare one week of your real workload.
  • IT admins. MXC is the piece that changes policy. It is how Windows plans to sandbox and audit AI agents on company PCs.
  • Everyone else. A normal laptop still runs ChatGPT, Claude and Grok Bot in a browser. RTX Spark only matters if you want models to run on the machine.

Human review risks

  • Calling October 7 the RTX Spark “launch” or “reveal.” It was announced in June. October 7 opened preorders.
  • Pairing $2,599 with 128 GB. The base model has less memory.
  • Presenting “matches cloud models” or “runs 120B locally” as tested.
  • Giving a Canadian price. None was published.
  • Making OpenAI, Anthropic, Mistral or Google the lead. Not done here.

Bottom Line

Believe the prices, dates and specs. They come from Microsoft and NVIDIA directly: Surface Laptop Ultra from $2,599, available October 16; Surface RTX Spark Dev Box $5,999, U.S. only, shipping in November; up to 6,144 GPU cores, a 20-core Grace CPU and 128 GB of unified memory. Believe that MXC is generally available. Microsoft says so. Treat the AI performance claims as marketing until reviewers test shipping units. Note that the 128 GB configuration is not the $2,599 one: retailers list the top Surface Laptop Ultra at about $5,899. DGX Station for Windows is a preview with no price or date.

Sources