NVIDIA’s RTX Spark laptops started shipping on October 16, 2026, putting a Grace Arm CPU and a Blackwell-based RTX GPU inside a single unified-memory chip and handing it to eight PC makers at once. The launch, detailed by NVIDIA and Microsoft, marks the company’s clearest attempt yet to turn “AI PC” from a marketing label into a laptop that can actually run big language models without touching the cloud.
Pricing starts at roughly $2,599 and climbs past $6,900 for the most loaded configurations, according to CNET’s live coverage of Microsoft’s October Windows and Surface event and a separate specs roundup from GadgetPilipinas. That price spread tells its own story: RTX Spark is not trying to be a mainstream laptop chip. It is trying to be a portable alternative to the AI workstations and small servers that developers and researchers currently rent by the hour.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What RTX Spark Actually Is: Grace CPU Meets Blackwell GPU
RTX Spark is not a traditional laptop processor paired with a separate discrete GPU. NVIDIA built it as a single superchip that fuses a Grace Arm CPU with a Blackwell-architecture RTX GPU, connected over NVIDIA’s own high-speed interconnect so the two halves share one pool of memory instead of splitting system RAM from dedicated VRAM. That design choice is the entire pitch: instead of a GPU running out of VRAM halfway through loading a large model, RTX Spark lets the CPU and GPU pull from the same, much larger memory pool.
NVIDIA frames the shift as a rethink of what a personal computer is for. Jensen Huang, founder and CEO of NVIDIA, said during the launch: “The PC is being reinvented,” according to NVIDIA’s official announcement. The comment reflects how NVIDIA is positioning RTX Spark less as a gaming chip and more as an on-device engine for AI agents that previously needed a data center connection to run at all.
Memory is where RTX Spark breaks from every prior laptop platform. NVIDIA says the chip uses shared LPDDR5X unified memory rather than separate system RAM and GPU VRAM pools, and that higher-end configurations can scale up to 128GB of that shared memory. For context on how tight AI memory supply already is heading into this launch, see our coverage of the HBM4 memory shortage driving AMD and its rivals to chase supply.
Two N1X Tiers: 18 Cores vs. 20 Cores, 24GB vs. 128GB
RTX Spark ships in two configurations of what NVIDIA calls the N1X chip. The entry tier pairs an 18-core CPU with a 5,120-core GPU, aimed at lighter laptop designs with up to 64GB of unified memory in some builds. The higher tier steps up to a 20-core CPU and a 6,144-core GPU, and it is this tier that unlocks the full 128GB memory ceiling in systems like ASUS’s workstation-class design.
The table below lays out how the two tiers compare, using the publicly disclosed Microsoft and ASUS configurations as real-world examples of where each tier lands on price.
| Spec | N1X Entry Tier | N1X Top Tier |
|---|---|---|
| Grace CPU cores | 18 | 20 |
| Blackwell GPU cores | 5,120 | 6,144 |
| Unified memory options | 24GB–64GB | Up to 128GB |
| NVIDIA-claimed AI compute | Not disclosed | Up to 1 petaflop |
| NVIDIA-claimed max context window | Not disclosed | Up to 1 million tokens |
| Example system (entry config) | Surface Laptop Ultra, 24GB/512GB – $2,599.99 | — |
| Example system (top config) | — | Surface Laptop Ultra, 128GB/1TB – $5,899.99 |
NVIDIA’s claim that the top N1X tier can run local AI models with as many as 120 billion parameters, and handle context windows up to one million tokens, is the number worth watching. If it holds up in independent testing, it would put a meaningfully large model within reach of a laptop battery rather than a rack of GPUs.
Pricing: $2,599 to Start, Nearly $7,000 at the Top
Microsoft’s Surface Laptop Ultra, the most fully documented RTX Spark system at launch, starts at $2,599.99 for an 18-core/5,120-core configuration with 24GB of unified memory and a 512GB SSD. Stepping up to 32GB of memory and a 1TB SSD pushes the price to roughly $3,300, and the fully loaded 128GB configuration with the 20-core/6,144-core chip and a 1TB SSD lists at $5,899.99, per CNET’s event coverage.
ASUS is pricing its ProArt P16 RTX Spark even higher: a 20-core, 128GB, 2TB configuration reaches $6,999.99, according to GadgetPilipinas’ specs breakdown. That places RTX Spark laptops well above typical premium ultrabooks and squarely in workstation territory, a pricing decision that mirrors how NVIDIA has priced its data-center-adjacent hardware for years rather than how it prices consumer GeForce laptops.
Eight Manufacturers Ship Simultaneously
NVIDIA says RTX Spark laptops are available from eight manufacturers at launch: Acer, ASUS, Dell, Gigabyte, HP, Lenovo, Microsoft, and MSI. That is an unusually broad simultaneous rollout for a brand-new chip platform, and it echoes the “literally every computer maker in the world” line Huang used to describe industry support for the launch, as quoted by Tom’s Hardware.
| Manufacturer | Known model | Starting price | Notes |
|---|---|---|---|
| Microsoft | Surface Laptop Ultra (15-inch) | $2,599.99 | Most detailed launch system |
| ASUS | ProArt P16 RTX Spark | Up to $6,999.99 (128GB/2TB) | Creator/workstation focus |
| Dell | Not disclosed at launch | Not disclosed | Confirmed launch partner |
| Lenovo | Not disclosed at launch | Not disclosed | Confirmed launch partner |
| HP | Not disclosed at launch | Not disclosed | Confirmed launch partner |
| MSI | Not disclosed at launch | Not disclosed | Confirmed launch partner |
| Acer | Not disclosed at launch | Not disclosed | Confirmed launch partner |
| Gigabyte | Not disclosed at launch | Not disclosed | Confirmed launch partner |
Several of those brands have not published exact RTX Spark model names or prices yet, which is normal for a staggered retail rollout but leaves buyers comparing Microsoft and ASUS pricing as the only firm reference points for now.
Why Microsoft Built Surface Laptop Ultra Around This Chip
Microsoft’s decision to lead with its own Surface hardware signals how central RTX Spark is to the company’s Windows AI strategy. Pavan Davuluri, Microsoft’s Executive Vice President for Windows and Devices, said: “We built Surface Laptop Ultra around NVIDIA RTX Spark,” according to NVIDIA’s blog recap of the joint Windows event. Davuluri also described the practical upside of the extra memory: “With up to 128 gigs of unified memory and up to a petaflop of AI compute, you can run models on this laptop that simply don’t fit on a traditional machine.”
That framing matters because Microsoft has spent two years pushing Copilot+ PC branding around on-device AI features that, in practice, have mostly meant small, specialized models running on modest NPUs. RTX Spark is a different scale of ambition: full-size language models that would otherwise require a cloud subscription or a dedicated workstation.
Running Large Models Locally: The 120-Billion-Parameter Claim
NVIDIA’s headline technical claim is that the top RTX Spark configuration can run AI models with up to 120 billion parameters entirely on-device, with support for context windows as large as one million tokens. Both figures come directly from NVIDIA’s own materials rather than independent benchmarks, so they should be read as the company’s stated ceiling rather than a verified, third-party-tested result.
Still, the scale is notable. Most laptop NPUs on the market today are built to handle small, efficient models measured in single-digit billions of parameters, not 120 billion. If RTX Spark’s unified memory architecture holds up under real workloads, it would close much of the gap between what a laptop can do locally and what previously required a rented GPU instance in the cloud, a shift that mirrors the broader AI infrastructure build-out we covered in NVIDIA B200 pricing and the Tencent-Oracle data center deal.
What NVIDIA and Microsoft Executives Are Saying
Beyond the specs, NVIDIA leaned heavily on Huang’s own enthusiasm to sell the platform’s ambition. Describing RTX Spark’s design during a press interview, Huang said: “That is the personal computer of the future. Tell me that’s not R2-D2. Tell me that’s not robotics. Tell me that’s not cool,” according to Tom’s Guide.
Huang also framed the launch in broader, almost historical terms, telling reporters: “We have a chance to reinvent the single most important instrument, the single most important tool of humanity,” per Tom’s Hardware’s interview coverage. That kind of language is standard for a major NVIDIA product launch, but it underlines how much weight the company is putting on RTX Spark succeeding as a category, not just a single product line.
Competitive Landscape: How RTX Spark Stacks Up
RTX Spark enters a crowded field of chips chasing the same “AI PC” category. AMD’s Ryzen AI Max (code-named Strix Halo) and Qualcomm’s Snapdragon X2 Elite both target on-device AI workloads with their own unified-memory-style designs, and Apple’s M-series chips have offered shared CPU/GPU memory pools in Macs for years. What sets RTX Spark apart on paper is that it pairs Grace CPU cores with a full Blackwell GPU architecture, the same architecture family NVIDIA sells into data centers, rather than a mobile-first GPU design scaled down for a laptop.
That distinction is also why RTX Spark is priced so differently from a typical AI-capable ultrabook. Rivals built around mobile CPU platforms generally top out well below $2,000 for their best configurations, while RTX Spark’s cheapest system starts where those competitors’ flagships end. For a sense of how volatile AI chip pricing has already gotten this year, see our report on NVIDIA GPU prices spiking 41% in Japan and whether the US market follows, plus how Microsoft’s own Maia 300 chip claims a 40% edge over Nvidia in data-center workloads, a separate but related front in the same AI-hardware race.
On the CPU side of the PC market more broadly, the pricing pressure is already visible. Our coverage of the 7800X3D vs. 9800X3D vs. 9850X3D price gap shows AMD pushing premium pricing across its own gaming CPU lineup even before factoring in AI-specific silicon like RTX Spark’s Grace cores.
Market Impact: A Premium Bet in a Shaky PC Market
RTX Spark arrives as the broader PC and smartphone hardware market is under real strain from memory shortages and rising component costs. Positioning laptops starting near $2,600, and running past $6,900 for top configurations, is a bet that enough buyers, developers, researchers, and AI-focused businesses, will pay a premium for local model inference rather than cloud compute rental. It is not a bet aimed at the mainstream laptop buyer.
That strategy lines up with how NVIDIA has approached previous compact AI hardware: price it like a specialized tool, not a consumer gadget, and let the capability justify the cost. Whether that translates into meaningful unit volume is an open question. None of the sources reviewed for this piece include independent analyst sales forecasts, stock-price reaction data, or unit shipment projections for RTX Spark, so any volume estimate at this stage would be speculation rather than reported fact.
Historical Context: From AI PC Marketing to Dedicated Silicon
The “AI PC” label has been applied loosely across the industry since 2024, often to laptops whose main qualification was a modest NPU bolted onto an otherwise conventional CPU. RTX Spark represents a sharper break from that pattern: rather than adding a small AI accelerator alongside existing CPU and GPU components, NVIDIA built a chip where the CPU and GPU share one memory pool from the ground up, explicitly to make large-model inference the laptop’s primary job rather than a side feature.
That shift mirrors a pattern playing out across the AI hardware industry more broadly, where companies are re-architecting chips specifically around memory bandwidth and capacity rather than raw compute alone, since memory has become the binding constraint for running larger models. It is the same dynamic behind this year’s unprecedented demand for high-bandwidth memory chips.
Compact RTX Spark Desktops Follow in November
NVIDIA has scheduled compact RTX Spark desktops for November 2026, a deliberate staggering rather than a simultaneous launch with the laptops. That sequencing suggests NVIDIA and its partners wanted laptop availability, and the retail infrastructure around it, locked in first before adding a second hardware form factor to the same chip platform. It also gives early laptop buyers and reviewers a few weeks to stress-test NVIDIA’s performance claims before a second wave of hardware arrives.
What This Means for Developers and AI Builders
For developers building or fine-tuning AI applications, RTX Spark’s pitch is straightforward: a 128GB unified memory pool means testing against a 120-billion-parameter model locally, without a cloud API bill or network latency, becomes realistic for the first time on a laptop. That could change workflows for teams currently paying for GPU cloud instances just to prototype against mid-size open models before deploying to production infrastructure.
It will not replace cloud infrastructure for training or large-scale serving. The use case NVIDIA is targeting is local development, inference testing, and running agentic AI workflows that benefit from low latency and data staying on-device, rather than competing with data-center-scale training clusters.
Predictions: Where RTX Spark Goes From Here
- Expect independent benchmarks within weeks of the October 16 launch to test whether the claimed 120-billion-parameter and one-million-token figures hold up outside NVIDIA’s own demos.
- Entry-level RTX Spark pricing will likely drift lower by mid-2027 as more of the eight launch manufacturers publish their own competing configurations and undercut Microsoft and ASUS on price.
- AMD and Qualcomm are likely to respond with their own unified-memory marketing pushes for Ryzen AI Max and Snapdragon X2 Elite successors, even without matching RTX Spark’s discrete-GPU-class architecture.
- The November 2026 compact desktop launch will probably undercut the cheapest RTX Spark laptop on price, targeting developers who don’t need portability but want the same memory ceiling.
- Expect scrutiny over real-world battery life and thermals, since pairing a data-center-derived GPU architecture with a laptop chassis is a genuinely new engineering trade-off NVIDIA has not had to solve at this scale before.
Frequently Asked Questions
What is NVIDIA RTX Spark?
RTX Spark is NVIDIA’s unified-memory AI PC platform that combines a Grace Arm CPU with a Blackwell-architecture RTX GPU on a single chip, sharing one pool of LPDDR5X memory instead of splitting system RAM from GPU VRAM.
When did RTX Spark laptops go on sale?
RTX Spark laptops began shipping on October 16, 2026, from eight manufacturers: Acer, ASUS, Dell, Gigabyte, HP, Lenovo, Microsoft, and MSI.
How much do RTX Spark laptops cost?
Pricing starts at $2,599.99 for Microsoft’s entry Surface Laptop Ultra configuration and rises to roughly $5,899.99 for Microsoft’s top configuration, with ASUS’s ProArt P16 RTX Spark reaching as high as $6,999.99 for its fully loaded 128GB, 2TB configuration.
What makes RTX Spark different from a regular gaming laptop chip?
Instead of a separate CPU and discrete GPU with their own memory pools, RTX Spark shares a single unified memory pool, up to 128GB, between CPU and GPU, which NVIDIA says allows laptops to run far larger AI models locally than a typical gaming laptop’s VRAM would allow.
Can RTX Spark run large AI models without the cloud?
NVIDIA claims the top RTX Spark configuration can run models with up to 120 billion parameters and context windows as large as one million tokens entirely on-device. These are NVIDIA’s own stated figures and have not yet been verified by independent third-party benchmarks.
When is the RTX Spark desktop version coming?
NVIDIA has scheduled compact RTX Spark desktops for November 2026, roughly a month after the laptop launch.
Is RTX Spark meant to replace a gaming GPU?
No. RTX Spark is positioned as an AI development and inference platform, not a gaming GPU replacement. Its pricing and target use cases, running large AI models locally, are aimed at developers, researchers, and AI-focused professionals rather than gamers.
Which laptop should buyers consider first at launch?
Microsoft’s Surface Laptop Ultra is the most fully documented RTX Spark system at launch, with four publicly listed memory configurations from 24GB to 128GB. ASUS’s ProArt P16 RTX Spark is the other model with confirmed specs and pricing, aimed more at creator and workstation workloads.
