NVIDIA announced a new hardware push for on‑device and on‑premise AI on Oct 23, 2026, unveiling a laptop-class RTX Spark product line and a DGX Spark system with a 64GB variant. The company positions the products to enable local personal agents and on‑device inference for frontier models.

What NVIDIA revealed

The RTX Spark laptops are described as laptop‑class systems offering up to 128GB of unified memory and delivering up to 1 petaflop of AI compute. NVIDIA presented these specifications as part of the product launch documentation published on Oct 23, 2026.

Alongside the laptop line, NVIDIA introduced a DGX Station branded for this initiative as DGX Spark, announcing a 64GB variant. The announcement also referenced larger DGX Station options intended for teams.

Positioning and intended use

NVIDIA framed the RTX Spark and DGX Spark systems as being suitable for “personal agents” and for running local or on‑prem inference, explicitly targeting workflows that put frontier models closer to end users or within enterprise premises rather than relying solely on remote cloud execution.

The company indicated day‑zero support for a selection of models in the launch material summarized in coverage of the announcement, signalling an intent to make the systems usable for a range of contemporary AI models at availability.

Availability and timelines

The technical details and launch information were published on Oct 23, 2026. NVIDIA’s materials include immediate launch messaging for some products and specify December deliveries for DGX Station Windows availability and certain systems, indicating a staggered availability across the announced portfolio.

Why this matters

The combination of high unified memory in a laptop form factor and petaflop‑scale compute is aimed at moving more inference and agent workloads from centralized cloud servers to local or on‑prem devices. This can shorten latency for interactive agents, reduce recurring cloud costs for some users, and provide organisations with physical control over data and model execution environments.

The DGX Spark 64GB option and the mention of larger team‑oriented DGX Station systems signal NVIDIA’s simultaneous focus on single‑user mobility and collaborative or small‑team on‑prem deployments.

Confirmed technical highlights

  • RTX Spark laptops: up to 128GB unified memory; up to 1 petaflop of AI compute.
  • DGX Spark: announced with a 64GB DGX Station variant; larger DGX Station options also noted for team use.
  • Launch publication date: Oct 23, 2026.
  • Availability: launch messaging indicates immediate availability for some products, with December deliveries specified for DGX Station Windows and certain systems.

The announcement materials present the products as ready to run frontier models locally and to support “personal agents,” although the launch coverage summarised which models would be supported at day‑one without reproducing a full model list in the published summary.

What remains unspecified in the announcement

The published launch material and coverage provide technical memory and compute figures and high‑level availability windows, but they do not list pricing, detailed performance benchmarks beyond the headline petaflop figure, nor a complete catalog of day‑one model support within the evidence provided.

NVIDIA’s launch message combines immediate availability language with explicit December delivery timing for some DGX Station configurations, so customers should expect a phased roll‑out rather than a single global ship date based on the disclosed material.

Further details beyond the published specifications and availability notes may be released by NVIDIA in subsequent product pages, press materials or through sales channels.