The Chip Is Real. Here's What It Actually Does.

After months of leaks and teaser posts from Nvidia, Microsoft, and ARM, Jensen Huang made it official at Computex 2026 on May 31, 2026. The RTX Spark is a complete ARM-based superchip — CPU, GPU, and memory on one package — built to power Windows laptops and compact desktops.

Unlike discrete graphics cards attached to someone else's processor, this is Nvidia's full-stack play to own the entire chip inside your machine.

The Specs — What We Actually Know

The flagship RTX Spark configuration ships with 20 ARM CPU cores, a Blackwell GPU with 6,144 CUDA cores, and 128GB of LPDDR5X unified memory at up to 300 GB/s of bandwidth, according to Nvidia's announcement and confirmation from Tom's Hardware. The CPU and GPU are connected via NVLink C2C — Nvidia's own high-bandwidth interconnect.

The chip is the same GB10 silicon that powers Nvidia's DGX Spark "personal AI supercomputer" mini-PC, which already sells to AI developers. Nvidia tuned it for Windows 11 and rebranded it as RTX Spark.

Lower-spec versions will follow — Nvidia told journalists the RTX Spark family will eventually offer as little as 16GB of RAM, targeting lower price points. No pricing announced yet for any tier.

Nvidia senior director of product management Mark Aevermann called it "the most efficient PC chip ever built." The Verge noted he provided no charts or statistics to support that claim.

Microsoft Is First Out the Door With Surface Laptop Ultra

Microsoft and Nvidia jointly announced the Surface Laptop Ultra — a 15-inch machine built around RTX Spark.

Microsoft Surface boss Andrew Hill told The Verge: "This is the most powerful thing we've ever made."

The hardware includes a 15-inch mini-LED touchscreen at 262 pixels per inch, 2,000 nits peak HDR brightness, the largest haptic trackpad Microsoft has ever put on a Surface, and ports including USB-C (appears to be three), USB-A, HDMI, a full-size SD card slot, and a headphone jack. Weight comes in under 4.5 pounds. Colors: dark grey and silver.

Pricing remains unannounced. USB speeds and versions are not disclosed. Final specs are still under revision. Microsoft's announcement blog was heavy on marketing language and light on hard numbers, per The Verge.

The Ghost of Surface RT Is Haunting This Launch

This moment echoes a previous chapter in tech history that deserves attention: the 2012 Surface RT.

Microsoft shipped the Surface RT as an ARM-based machine powered by Nvidia's Tegra chip. It failed. Microsoft wrote off $900 million on that bet. Apps didn't work. Performance suffered. The platform died.

Now, in 2026, Nvidia and Microsoft are back with the same basic pitch: ARM-based Windows machine.

The difference is that Microsoft has spent years refining its Prism emulation layer to run legacy x86 software on ARM chips — work initially done for Qualcomm's Snapdragon X Elite platform. Nvidia is inheriting and extending that foundation. x86 emulation still carries a performance penalty, however, and that gap matters for professionals running legacy software stacks.

Who Else Is Building RTX Spark Machines?

Microsoft isn't alone. According to Tom's Hardware, Nvidia confirmed RTX Spark-powered laptops from Dell, HP, Lenovo, Asus, and MSI are in development. Every major Windows OEM is committed.

This is a coordinated platform launch, not a one-off experiment.

DLSS 4.5 and the Gamer Angle

Nvidia also announced at Computex that DLSS 4.5 Ray Reconstruction arrives in August 2026, with Blender support in Fall 2026, per Nvidia's GeForce blog. Over 1,000 RTX-enhanced games and apps are available. Another 11 games are adding DLSS 4.5 support, including Marvel Rivals and Gothic 1 Remake.

The Nvidia App update adds 240 FPS ShadowPlay recording for capture work.

For gamers and creators, these are substantive additions — overshadowed by Nvidia's emphasis on AI agent narratives.

What the AI Pitch Actually Means

Nvidia is selling a vision of the PC as an agentic AI platform — where software agents run tasks autonomously using local models up to 120 billion parameters with context lengths of up to a million tokens, per Nvidia's specifications.

This justifies 128GB of unified memory in a laptop. Most consumers won't need it. Enterprise buyers, developers, and AI researchers running local models have a legitimate use case.

Nvidia's new OpenShell runtime for Windows promises 2x inference performance on agentic models via llama.cpp and vLLM. Adobe and Blender are already adapting their apps for RTX Spark, according to Nvidia.

The Real Question

Nvidia just declared competition against Apple Silicon, Qualcomm Snapdragon X, Intel Core Ultra, and AMD Ryzen AI — simultaneously — with a chip that has no published independent benchmarks.

A company claiming "most efficient ever" without supporting data deserves scrutiny. Surface Laptop Ultra ships "this fall." The actual test — performance numbers, battery life, x86 app compatibility, and price — comes later.

Wait for the benchmarks.