NVIDIA says RTX Spark laptops became available on October 16, moving the Windows machines from the preorders announced at Microsoft's October 7 event into a retail launch. The company's availability release names models from ASUS, Dell, HP, Lenovo, Microsoft and MSI. Its October 17 post reports launch events rather than a second hardware release date.

For someone buying a machine to run local AI, the important questions are which configuration is being sold in their region and which software is ready to use. RTX Spark is a chip platform, not one laptop specification. NVIDIA advertises up to 128 GB of shared CPU and GPU memory and up to one petaflop of FP4 AI compute; an entry model need not have either maximum. BIG CHANGE has not tested the laptops or independently measured their performance.

The big change

  • What changed: NVIDIA now describes RTX Spark laptops from six PC makers as available, after the October 7 preorder announcement. The October 17 blog documents retail celebrations of that launch.
  • Who it affects: Developers and creators considering local models or accelerated media work can compare named models and their actual memory and software configurations. Regional stock and delivery dates still require a check at the seller.
  • What follows: GitHub already offers a public-preview route to select a locally supplied model in the Copilot app, and HydraFusion is available in research preview. Microsoft dated its plan for automatic local and cloud routing to later October. Check the precise client and feature before planning a workflow around it.

The laptops and the price need a configuration check

NVIDIA's release lists 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 Flip AI+ and Prestige N16. It says the devices are available globally through the makers and major retailers, with prices and configurations varying by region. Individual seller pages can still show a preorder or an out-of-stock option; NVIDIA's platform announcement is not a stock guarantee for every model.

The two NVIDIA launch pages disagree on the entry price. The October 16 newsroom release says laptops start at $2,599; the October 17 blog says $2,399. Neither figure identifies a 128 GB machine. A search-index snapshot of Microsoft's US Surface Laptop Ultra configuration page showed a 24 GB entry option and 128 GB as a separate option. We could not load the live store page to confirm its current price or stock. A buyer should use the selected seller's memory, processor, price and delivery listing for a purchase decision.

The shared-memory design matters to local models because the CPU and GPU draw from one pool. NVIDIA says the platform can fit larger local models and creative scenes on configurations with sufficient memory. It has not established that every listed laptop, or the least expensive version, can run a given model at a useful speed. Its figures for frame rates, rendering and model capacity are company claims under its own conditions, not measurements made for this article.

Software arrives on separate schedules

NVIDIA's October 13 software roundup identifies several different release stages. It says NVIDIA PAIR, a local-network inference router, moved out of beta, and NVIDIA Broadcast 2.3 is available for download. RTX Spark optimizations for Photoshop and Premiere are in beta, with Substance 3D updates coming later. The same roundup says Hermes Agent's granular OpenShell controls and Microsoft Execution Containers integration are due by the end of October. Those are not all day-one, final releases.

Our earlier Windows launch report separated the same hardware preorder from Microsoft's software timetable. Microsoft's Execution Containers are generally available for developers. GitHub's current Copilot app documentation describes public-preview support for adding a local model provider and selecting its model in an agent session. GitHub's October 14 changelog says HydraFusion's multi-model orchestration is available in research preview across VS Code, the Copilot app and CLI, subject to plan and preview settings.

The October 7 GitHub and Microsoft explanation dated automatic routing between on-device and cloud inference for later that month. NVIDIA's October 13 roundup says Copilot has integrated its own routing solution and points to local setup, but it does not specify whether that exact automatic local/cloud mode is released to every client. Broader Copilot context and action features have a separate rollout on eligible Copilot+ PCs, with device and market conditions. RTX Spark hardware availability alone does not establish a release date for any one of these software options.

The practical next step is to compare an orderable regional configuration with the memory required by a specific model or creative workload, then check the current release channel of the app that will use it. This is a hardware availability update, not a hands-on verdict on whether the complete local AI workflow performs as advertised.

Sources & further reading