What is Nvidia RTX Spark?

It's a new superchip for laptops and desktops combining a Blackwell RTX GPU with 6,144 CUDA cores, a 20-core Grace CPU (MediaTek), 128GB of unified memory, and 70 billion transistors on TSMC's 3nm process. It delivers 1 petaflop of AI performance in a portable laptop form factor.

What's the difference between RTX Spark and Nvidia's previous chips?

RTX Spark combines GPU, CPU, and memory into one cohesive system with unified memory — the GPU and CPU share the same memory pool, reducing data-transfer latency. It's specifically optimized for AI agents and on-device workloads.

Will RTX Spark be available in consumer laptops?

Yes. Nvidia is building a line of PCs with Microsoft that are 100% Windows compatible, with full CUDA support. Laptops and desktops with RTX Spark will arrive in fall 2026.

What is Vera Rubin and what does it do?

Vera Rubin is Nvidia's multi-rack platform for AI data centers handling agentic workloads. It's now in full production. Jensen Huang called it 'the most ambitious project in Nvidia's history,' designed entirely on TSMC's 3nm process.

When will PCs with RTX Spark arrive in Italy?

No precise date has been announced for Italy, but the PCs should arrive globally in fall 2026. Nvidia's partners (and Microsoft) will announce local versions in the coming months.

Nvidia RTX Spark and Vera Rubin GTC 2026 — AI on Laptops and in Data Centers

Nvidia RTX Spark: Portable AI Arrives with a New Superchip

TL;DR: At GTC Taipei 2026, Jensen Huang unveiled RTX Spark: a new Nvidia superchip combining a Blackwell GPU, a Grace CPU (MediaTek), and 128GB of unified memory. 1 petaflop of AI performance in a laptop. Vera Rubin, Nvidia's platform for AI data centers, is now in full production. RTX Spark PCs will arrive in fall 2026, Windows compatible, with full CUDA support.

AI will no longer need to be shipped online for processing. With RTX Spark, AI lives in your laptop.

RTX Spark: The Chip That Brings AI to the Laptop

RTX Spark is the result of three years of engineering focused on one question: how do we fit a petaflop of AI performance into a portable form factor while keeping power consumption manageable?

Here are the specs:

Component	Specs
GPU	Blackwell RTX with 6,144 CUDA cores
CPU	20-core Grace (co-designed with MediaTek)
Memory	128GB unified LPDDR6
Process	TSMC 3nm
Transistors	70 billion
AI Performance	1 petaflop

Unified memory is the key. Unlike traditional chips where GPU and CPU have separate memory, in RTX Spark the GPU and CPU share the same 128GB pool — zero copying between discrete memory. For AI workloads, this means lower latency and higher throughput.

The Vision: "Useful AI Has Arrived"

Jensen Huang opened the keynote with a blunt statement: "Useful AI has arrived." Not demos, not research, not lab experiments. Useful AI — systems running 24/7, agents completing tasks, models solving real problems.

He then added: "AI is now a profit generator." For Nvidia, this means it's no longer just selling chips for research training. It's selling infrastructure that makes money for its customers.

RTX Spark is this vision materialized in a consumer laptop. A device that runs local AI models with no connection, no cloud latency, no API costs.

Vera Rubin: AI Data Centers in Full Production

While RTX Spark brings AI to the laptop, Vera Rubin scales AI up to the data center level.

Vera Rubin isn't a single chip — it's a complete multi-rack platform, designed entirely by Nvidia for agentic workloads and AI inference at scale. It's now in full production, with 40,000 Nvidia engineers involved in its development.

Jensen called it "the most ambitious project in Nvidia's history" — a statement that carries weight given we're talking about the creators of GPU computing.

Vera Rubin is entirely built on TSMC's 3nm process. It's the opposite of RTX Spark (portable) — it's gigantic, built for cloud providers and companies running millions of AI agents in parallel.

Windows PCs: The New Era of the Laptop

Nvidia is co-developing a new line of PCs with Microsoft, built from the ground up for on-device AI.

Specs:

OS: Windows 11 (not a custom Linux build)
Compatibility: 100% Windows compatible, full CUDA support
Form factor: Laptop, desktop, workstation
Availability: Fall 2026

It's a significant move. Nvidia — which has always left laptop design to OEM partners (HP, Dell, Lenovo) — is entering the PC market directly with its own line. Similar to how Intel handled reference designs, but Nvidia is moving more aggressively.

The Context: AI Is No Longer Just a Cloud Service

The underlying message of GTC 2026 is that AI is moving back on-device.

For 18 months, the paradigm was: run your model in the cloud, pay for the API, accept the latency and the cost. It worked for many use cases. But the constraints are mounting: privacy (data doesn't leave the device), latency (25 milliseconds vs. 500ms in the cloud), cost (1,000 queries a month adds up).

RTX Spark's pitch: load the model onto your laptop, run it locally, no cloud, no per-query cost. For agentic AI (systems that reason and act in the background), it's a game-changer.

Conclusion

RTX Spark and Vera Rubin represent the same strategy at two different scales: making AI omnipresent — both in your pocket and in the data center. Nvidia no longer just sells chips: it sells the infrastructure for a generation where AI runs everywhere, always, with no cloud dependency.

Fall 2026 for the PCs, today for Vera Rubin in data centers. Nvidia is moving fast.

Sources: Nvidia GTC Taipei 2026 Keynote, The Next Web — Jensen Huang Computex 2026, TechRadar — Nvidia Computex 2026, Tom's Hardware — RTX Spark Roadmap