RTX Spark laptops ship from $2,599: pilot local AI on one machine

Nvidia RTX Spark laptops are on sale today, from six PC makers, starting at $2,599. Nvidia announced global availability on October 16. The same day, JetBrains put its free Junie Local coding agent on RTX Spark. Local AI on a Windows laptop is no longer a preorder. It is a purchase order. Our advice: buy one, not ten.

What actually happened

Per Nvidia:

• The machines: ASUS ProArt P14 and P16, Dell XPS 16, HP OmniBook Ultra and OmniBook X 14, Lenovo Yoga Pro 9n, Microsoft Surface Laptop Ultra, and MSI Prestige N16 models.
• The chip: One superchip with a Blackwell RTX GPU and a Grace CPU, built with MediaTek. Up to 128GB of unified memory and up to 1 petaflop of FP4 AI performance.
• The software: Nvidia names GitHub Copilot, Perplexity, OpenClaw and Hermes among the agents, plus its OpenShell runtime and Microsoft Execution Containers for sandboxing.
• The model: Nemotron 3.5 Bolt, a 74-billion-parameter model tuned for tool calling, on supported configurations.
• The price: From $2,599. Nvidia says pricing and configurations vary by region.

Per JetBrains, Junie Local is free, with no subscription or credits. It runs Qwen3.6-27B on your machine, reads your code, edits files and runs tests. It needs at least 24GB of GPU-addressable memory and about 20GB of disk. JetBrains calls its 55 tokens per second an "initial performance baseline," and says your code never leaves the machine.

Why it matters for your business

When Microsoft priced the Surface hardware, we said to price local AI against your API bill. That math hasn't changed. What changed is that you can now test it this week on hardware from your usual vendor.

Run a pilot before you change a fleet:

• Pick one workflow with sensitive data. Client code, contracts, customer records. Local AI earns its price where data can't go to a cloud model.
• Buy the memory, not the badge. Context size depends on GPU memory. A base configuration may not fit the model you want.
• Run it side by side for two weeks. Same tasks, local and cloud. Track time saved and how many outputs need rework.
• Count the hidden costs. Driver updates, model downloads, and someone who owns the setup.

If local wins on one machine, expand. If it doesn't, you spent $2,599 to learn that, not $26,000.

Sources: Nvidia Newsroom, JetBrains Junie.