Nvidia begins selling RTX Spark laptops starting at USD $2,599

By Marcus Bennett

Nvidia began selling RTX Spark laptops on November 13, 2026, with prices starting at USD $2,599 through Asus, Dell, HP, Lenovo, Microsoft and MSI.

The launch marks the commercial arrival of a product line Nvidia has used to move AI processing from the data centre toward personal computers. It also puts that strategy into premium Windows retail, where the company can test demand for local AI, gaming and creator features at prices above USD $2,500.

For buyers, Nvidia's proposition combines local AI, graphics performance and portability in the same thin-and-light machine. For the company, that creates one retail test spanning software development, content creation and gaming rather than three separate product claims.

Nvidia said 128GB configurations are already on retail shelves.

What hardware ships inside the RTX Spark laptops?

Nvidia says the systems combine an Nvidia Blackwell RTX graphics processor and an Nvidia Grace central processor in a single-chip design for thin-and-light Windows laptops.

• Nvidia says the laptops offer up to 128GB of unified memory, with graphics, AI models and other workloads sharing one memory pool.
• Nvidia says they deliver up to 1 petaflop of FP4 (four-bit floating-point) AI performance.
• Nvidia says configurations are as thin as 14mm and weigh as little as three pounds.
• Nvidia targets software development, content creation and gaming on portable machines.

The 128GB and 1-petaflop figures describe separate parts of Nvidia's platform: shared memory capacity and FP4 AI throughput. The company presents both inside a portable design aimed at developers, creators and gamers. Demand for that combined package is the commercial test Nvidia set for the retail release.

What can Nemotron 3.5 Transcribe do?

Nvidia introduced Nemotron 3.5 Transcribe as an open-weight speech model for transcription, speaker attribution, translation and audio question answering in one system.

Nvidia says developers can use plain-language prompts to transcribe a meeting, identify speakers, translate audio into English or extract decisions from a conversation. Prompts can also include industry-specific terms and formatting rules, which Nvidia says could help users in healthcare, financial services and customer service.

The model supports 40 languages, including seven Indic languages. Nvidia said Nemotron 3.5 Transcribe ranks first on the Hugging Face Open ASR leaderboard and second among open-weight models on Artificial Analysis.

The company said the model uses about 4.3 billion active parameters, 80% fewer than the top-ranked open-weight rival. Open weights let developers choose where audio is processed, whether on a desktop machine or in the cloud.

Nvidia says quantised (reduced-precision) versions are due to run locally on RTX Spark and DGX Spark systems, Jetson AGX Thor devices and Blackwell data centre graphics processors. Its deployment range therefore covers portable PCs, developer systems, embedded hardware and data-centre GPUs.

How can users manage competing GPU workloads?

Another Nvidia App update gives users direct control over how background AI agents and foreground applications share graphics processing resources. The new setting appears in the advanced system section of Nvidia App.

Nvidia describes the option as letting users “prioritise either the active application or long-running background tasks.” The feature addresses a practical issue for AI-enabled personal computers, where multiple workloads can compete for the same processor resources.

Nvidia says the balance can affect both responsiveness and the speed of AI tasks running in parallel on laptops used for editing, development or gaming. The setting turns contention between immediate work and long-running agents into a direct scheduling choice.

What does Liquid AI's LFM3 add?

Nvidia is also making Liquid AI's LFM3 model family available on RTX Spark through the llama.cpp and vLLM runtimes. Nvidia said the models cover text, vision, audio and decision tasks and “can be fine-tuned locally.”

The company lists several local workflows:

• Inbox triage, calendar management, document reading and screen analysis.
• Speech transcription, spoken question answering and support request routing.

Nvidia said the latest 1.7-billion-parameter base model outperforms Liquid AI's previous 2.6-billion-parameter version. The smaller release therefore combines a lower stated parameter count with higher claimed performance in Nvidia's comparison.

The listed workflows extend from reading documents and screens to processing speech and routing support requests. Combined with support for four task types and local fine-tuning, they give developers multiple ways to build local agents without separating model adaptation from deployment.

Which gaming updates accompany the launch?

Nvidia tied the laptop availability announcement to several gaming updates intended to showcase the machines' graphics features. Phantom Blade Zero launches with ray tracing and Nvidia DLSS 4.5 support, according to Nvidia, following work with developer and publisher S-GAME.

For Minecraft Java Edition 1.21.1, RTX Remix 1.6 adds path tracing (a ray-based rendering method) and DLSS support. Nvidia said the update also brings native RTX Spark support, new rendering features and workflow changes for modders using Portal with RTX and the Half-Life 2 RTX demo.

The Minecraft update is tied to a specific game version and RTX Remix release. Nvidia is using those titles to make the laptops' gaming features explicit at launch, alongside its broader positioning for developers and content creators.

What other local-AI products are available?

Outside the laptop range, DGX Spark 64GB is now available through manufacturing partners. It uses the GB10 Grace Blackwell Superchip and targets developers and researchers building and running AI models locally.

Nvidia also said Nemotron 3.5 Bolt E4B 4GB is now available for local voice assistants, game characters and embedded interfaces. A larger Nemotron 3.5 Bolt 74B-A9B 24GB model is intended for more demanding agent workloads on RTX Spark.

Taken together, the announcements extend Nvidia's local-model push across portable PCs, developer systems, Jetson AGX Thor devices and data-centre GPUs. Nvidia positions the 4GB E4B for voice assistants, game characters and embedded interfaces, while the 24GB 74B-A9B targets heavier agent work on RTX Spark.

RTX Spark laptops are shipping now, with 128GB configurations already on retail shelves, Nvidia said. The immediate test is commercial: whether demand spans all three audiences Nvidia named—software developers, content creators and gamers—at prices starting at USD $2,599 for Windows laptops that run speech and agent models locally while also serving as gaming and creator systems.