NVIDIA used its Computex 2026 stage to put a desktop class AI engine into the body of a thin laptop. The NVIDIA RTX Spark Superchip pairs an Arm CPU with a Blackwell GPU and a large shared memory pool on a single package, built with Microsoft as a Windows on Arm platform aimed squarely at running big AI models locally rather than streaming them from a data centre.

Quick Answer

The RTX Spark Superchip combines up to 20 Arm CPU cores with a Blackwell GPU carrying 6,144 CUDA cores and 128GB of unified LPDDR5X memory, all linked by NVIDIA's NVLink-C2C interconnect at up to 300 GB/s. Announced on 31 May 2026 at GTC Taipei, it powers a new class of Windows on Arm laptops and compact desktops arriving from major brands in late 2026.

What Actually Sits on the Chip

The headline is integration. Instead of a separate CPU, GPU, and pools of memory talking over slow links, Spark places an Arm CPU of up to 20 cores and a Blackwell GPU on one package and connects them with NVLink-C2C. Both halves then share a single 128GB block of LPDDR5X rather than each owning its own walled off memory.

That shared memory is the point. A traditional laptop GPU might have 8 or 16GB of dedicated VRAM, and anything larger has to spill into system RAM over a slow bus. With 128GB unified and up to 300 GB/s of bandwidth, the GPU can reach far more memory directly, which is what lets a portable machine hold a very large model in the first place.

Why Windows on Arm Matters Here

Arm chips have long led on efficiency, which is why phones use them, but Windows on Arm struggled because much desktop software was built for x86. NVIDIA and Microsoft are betting that pairing Arm efficiency with a serious Blackwell GPU changes that equation: you get long battery life from the Arm cores and heavyweight AI throughput from the GPU in the same chassis.

NVIDIA is framing Spark as the engine for an agentic Windows, where the operating system runs AI assistants and large models on device. The memory pool is sized for models in the 120 billion parameter range and context lengths reaching into the millions of tokens, the kind of long running local task that previously meant a desktop workstation or a cloud bill.

What It Means for Buyers

Spark laptops and small desktops are expected from ASUS, Dell, HP, Lenovo, Microsoft, and MSI in the second half of 2026, with Acer and Gigabyte to follow, and Microsoft's own Surface Laptop Ultra among the first. NVIDIA has also outlined a roadmap beyond this generation, with future Spark chips tied to its Rubin and later architectures, so this is a platform rather than a one off product.

For South African buyers the practical questions will be local availability, price in Rand, and which workloads genuinely benefit from on device AI versus a strong conventional GPU. If you want to compare the on device AI direction against current desktop and laptop options, the AI PC range at Evetech is the starting point, and the current GPU best sellers show what local creators are buying for AI work today while the Spark wave lands.

Frequently Asked Questions

What is the RTX Spark Superchip in one sentence?

It is a single package combining an Arm CPU of up to 20 cores, a Blackwell GPU, and 128GB of unified memory, built by NVIDIA and Microsoft as a high performance Windows on Arm platform for on device AI.

How much memory does it have and why does that matter?

128GB of LPDDR5X is shared between the CPU and GPU at up to 300 GB/s. A shared pool that large lets a portable machine hold very large AI models directly, instead of being capped by a small block of dedicated VRAM.

When can you actually buy a Spark device?

NVIDIA said Spark laptops and compact desktops would arrive in late 2026 from brands including ASUS, Dell, HP, Lenovo, Microsoft, and MSI, with more to follow. South African timing and pricing will depend on local importers.

Is Spark a gaming chip or an AI chip?

Its design centres on AI: a large unified memory pool and a Blackwell GPU tuned for running big models and agents locally. It can drive graphics, but the platform's headline purpose is on device AI rather than top tier gaming frame rates.

Want a machine built for local AI right now? Explore the AI PC lineup at Evetech and see how today's GPUs handle the AI workloads the Spark platform is being designed around.