NVIDIA’s RTX Spark laptops are officially on sale as of October 16, 2026, and the pitch is unusual for a Windows PC: a single chip fuses a Blackwell RTX GPU with NVIDIA’s own Grace CPU, the same Arm-based processor family NVIDIA built for data center servers. Six of the biggest PC brands, ASUS, Dell, HP, Lenovo, Microsoft, and MSI, are shipping machines built around it, with prices starting at $2,599, according to Tom’s Hardware.
That price point and the Grace-plus-Blackwell combination mark a real shift for NVIDIA. The company has spent the past two years selling AI muscle almost exclusively to data centers and through its DGX Spark desktop appliance. Now that same superchip architecture is landing in laptops consumers can buy at retail, aimed squarely at developers, researchers, and creative professionals who want to run large AI models locally instead of renting cloud GPU time. This is the top hardware story in tech news today, and it carries real implications for how the PC industry prices and positions AI capability going forward.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
NVIDIA RTX Spark Laptops Go on Sale at $2,599
NVIDIA confirmed general availability for RTX Spark laptops on October 16, 2026, with the lineup landing across six major OEM partners at once rather than trickling out brand by brand. The entry price of $2,599 applies to multiple models rather than a single flagship, which is notable because NVIDIA’s previous local-AI hardware, the DGX Spark desktop unit, carried a far steeper price tag aimed at enterprise buyers and research labs.
Microsoft’s Surface Laptop Ultra leads the pack at $2,599.99, putting a Microsoft-branded machine at the front of a launch built on NVIDIA silicon, an arrangement that would have seemed unlikely a few years ago when Microsoft leaned almost entirely on Intel and Qualcomm for Surface hardware. ASUS matches that price with its ProArt P14 RTX Spark model, while stepping up to the larger ProArt P16 RTX Spark pushes the price to $4,399.99. Dell, HP, Lenovo, and MSI are also listed as launch partners, according to PCMag, though NVIDIA’s own retail rollout varies by region and retailer.
For a market that has grown accustomed to AI laptop marketing meaning little more than a slightly faster neural processing unit bolted onto a conventional chip, RTX Spark laptops represent something structurally different. The GPU and CPU here are not separate components wired together on a motherboard. They are part of the same superchip package, unified by shared memory, a design NVIDIA first proved out in its data center Grace Hopper and Grace Blackwell platforms before shrinking the concept down to a laptop chassis.
What’s Inside the RTX Spark N1X: Blackwell GPU Meets Grace CPU
The flagship configuration NVIDIA lists is the RTX Spark N1X, which pairs a 6,144-core NVIDIA Blackwell RTX GPU with a 20-core NVIDIA Grace CPU, according to NVIDIA’s own RTX Spark specifications page. That Grace CPU is the same Arm-based processor lineage NVIDIA developed for server-class workloads, not a repurposed mobile chip, which is part of why the pairing has drawn attention from hardware reviewers used to seeing Intel or AMD silicon paired with NVIDIA graphics rather than NVIDIA building both halves of the equation itself.
The GPU side brings fourth-generation ray-tracing cores and fifth-generation Tensor Cores, the same core architecture family found in NVIDIA’s current Blackwell data center and desktop GPUs, scaled down for a laptop thermal envelope. NVIDIA rates the top-end configuration at up to 1 petaflop of AI performance, a figure aimed squarely at developers who want to run inference and fine-tuning jobs on large language models without shipping data out to a cloud provider.
What ties the GPU and CPU together, and what NVIDIA is leaning on hardest in its marketing, is unified memory. Rather than splitting system RAM from dedicated video memory the way almost every Windows laptop has worked for decades, RTX Spark machines share a single memory pool between the CPU and GPU. On the flagship configuration, that pool scales up to 128GB of LPDDR5X, which NVIDIA and reviewers both frame as the headline differentiator against conventional discrete-GPU laptops, where GPU memory is capped far lower regardless of how much system RAM is installed.
Why unified memory matters for local AI
Running a large language model locally is largely a memory capacity problem before it is a raw compute problem. A model with tens of billions of parameters needs enough addressable memory to hold its weights, and consumer GPUs have traditionally topped out well short of what serious local AI work demands. By giving the CPU and GPU shared access to up to 128GB, NVIDIA is betting that developers will pay a premium for the ability to load bigger models locally rather than being forced into cloud inference or aggressive quantization just to fit a model into a smaller VRAM pool.
Full Spec Breakdown: 6144-Core vs. 5120-Core Configurations
NVIDIA is shipping RTX Spark laptops in two distinct silicon configurations rather than one chip with memory options bolted on top. The flagship pairs the 6,144-core Blackwell GPU with the 20-core Grace CPU and scales up to 128GB of unified LPDDR5X memory. A second, lower-tier configuration steps down to a 5,120-core Blackwell GPU and an 18-core Grace CPU, with unified memory capped at 32GB.
That is a meaningful gap, not a minor SKU variation. Buyers choosing the 5,120-core, 32GB configuration are getting roughly a quarter less GPU core count and a quarter of the memory ceiling of the flagship, which will matter a great deal for anyone planning to run memory-hungry local AI workloads rather than general productivity tasks. Both configurations run Windows 11 and share the same fourth-generation ray-tracing cores and fifth-generation Tensor Core architecture, and both operate within a 45 to 80 watt TDP range, according to NVIDIA’s specifications page.
What the TDP range means for battery life and thermals
A 45 to 80 watt TDP window is wide enough that laptop makers have real latitude in how they tune individual models. OEMs chasing thinner, quieter chassis designs can clock closer to the 45 watt floor, trading peak performance for battery life and lower fan noise, while vendors targeting workstation-class buyers can push toward 80 watts for sustained AI workloads at the cost of heavier cooling hardware and shorter unplugged runtime. NVIDIA has not published per-model TDP targets for each OEM’s laptop, so actual sustained performance will vary by brand and chassis design once independent reviewers get hands-on time with specific units.
RTX Spark Pricing and Configurations at a Glance
| Model | Maker | Screen Size | Unified Memory | Starting Price |
|---|---|---|---|---|
| Surface Laptop Ultra | Microsoft | 15-inch | Up to 128GB | $2,599.99 |
| ProArt P14 (RTX Spark) | ASUS | 14-inch class | Up to 128GB | $2,599.99 |
| ProArt P16 RTX Spark | ASUS | 16-inch | Up to 128GB | $4,399.99 |
| RTX Spark laptop (various) | Dell, HP, Lenovo, MSI | Varies by model | Varies by configuration | Starting at $2,599 |
The spread between the ASUS ProArt P14 and ProArt P16 is instructive. Both carry the ProArt branding ASUS reserves for its creator-focused lineup, yet moving from the 14-inch class model to the 16-inch model adds $1,800 to the price, a jump that likely reflects a larger display, bigger battery, and almost certainly a cooling system built to sustain the chip’s upper TDP range for longer stretches rather than a difference in the core silicon itself.
Which Laptop Makers Are Shipping RTX Spark Hardware
Six OEMs are confirmed at launch: ASUS, Dell, HP, Lenovo, Microsoft, and MSI. That is a broad simultaneous rollout for a new chip platform, closer to how NVIDIA has historically launched new GeForce GPU generations across the whole PC industry at once rather than how niche AI-focused hardware has typically reached the market through one or two boutique vendors first.
HP is already listing RTX Spark machines through its own retail storefront, a sign the laptops are moving through standard consumer sales channels rather than being gated behind developer preview programs or enterprise procurement, as NVIDIA’s DGX Spark desktop largely was. That matters for how quickly the platform can build a user base. A chip confined to enterprise sales channels takes years to build developer mindshare; a chip sold next to ordinary consumer laptops at Best Buy and through OEM websites reaches a far wider audience far faster, which appears to be exactly the distribution strategy NVIDIA is pursuing here.
Microsoft’s involvement is particularly notable given the Surface Laptop Ultra’s positioning. Microsoft has used the Surface Laptop line as a showcase for whichever silicon platform it wants the broader Windows ecosystem to rally behind, and placing NVIDIA’s superchip front and center signals that Microsoft sees local AI capability, not just raw CPU or graphics performance, as the next real differentiator for premium Windows hardware.
Why NVIDIA Paired a Blackwell GPU With an Arm-Based Grace CPU
NVIDIA has sold GPUs to run alongside other companies’ CPUs for its entire history as a company. Grace changes that relationship. Rather than ceding the CPU half of the laptop equation to Intel, AMD, or Qualcomm, NVIDIA designed its own Arm-based CPU specifically to sit on the same package as its own GPU, with memory architecture built from the ground up for the two to share data without the latency and bandwidth penalties of a traditional CPU-GPU link over a PCIe bus.
That design choice has a strategic dimension beyond pure engineering. By controlling both halves of the chip, NVIDIA captures more of the laptop’s total silicon value rather than selling a discrete GPU module to be paired with someone else’s processor. It also lets NVIDIA tune the CPU specifically around feeding its own GPU and Tensor Cores rather than optimizing for general-purpose workloads the way Intel and AMD’s laptop chips have to. The tradeoff is that RTX Spark laptops run on Arm-based Windows rather than the x86 architecture that still underpins the vast majority of Windows software, which raises real questions about application compatibility that NVIDIA has not fully addressed in its public materials so far.
The Local AI Angle: 1 Petaflop in a Laptop Chassis
The up-to-1-petaflop AI performance figure NVIDIA attaches to the flagship RTX Spark N1X configuration is the number the company most wants developers to notice. A petaflop of AI throughput inside a laptop chassis would have sounded like a server rack specification just a few years ago, and NVIDIA is clearly positioning RTX Spark as a portable extension of the same local-AI pitch it built around the DGX Spark desktop unit earlier in 2026.
The practical case NVIDIA is making to developers is straightforward: cloud GPU rental for model fine-tuning and inference carries ongoing cost and, for some organizations, data residency concerns that a locally owned machine sidesteps entirely. A developer who buys an RTX Spark laptop once and runs inference workloads on-device avoids both the recurring bill and the need to ship proprietary data to a third-party server, an argument that resonates particularly strongly with enterprise AI teams working under strict compliance requirements.
Whether that pitch holds up against renting cloud GPU capacity depends heavily on workload. Short, bursty inference jobs may genuinely be cheaper and faster locally. Training runs that need to scale across dozens of GPUs simultaneously will still require cloud infrastructure or a dedicated on-premises cluster no single laptop, however capable, can replace.
RTX Spark Technical Specifications by Configuration
| Spec | RTX Spark N1X (Flagship) | RTX Spark (Lower Tier) |
|---|---|---|
| GPU | 6,144-core Blackwell RTX | 5,120-core Blackwell RTX |
| CPU | 20-core Grace CPU | 18-core Grace CPU |
| Unified Memory | Up to 128GB LPDDR5X | Up to 32GB LPDDR5X |
| AI Performance | Up to 1 petaflop | Not disclosed by NVIDIA |
| TDP | 45-80W | 45-80W |
| Ray-Tracing Cores | 4th-generation | 4th-generation |
| Tensor Cores | 5th-generation | 5th-generation |
| Operating System | Windows 11 | Windows 11 |
The four-times memory gap between the two configurations, 128GB versus 32GB, is the number buyers should weigh most carefully before choosing a tier. For anyone running general productivity software or gaming, the lower-tier configuration’s 32GB ceiling is unlikely to be a bottleneck. For developers specifically chasing local large language model inference, that same 32GB can be the difference between loading a mid-sized open-weight model comfortably and hitting a hard memory wall well short of what a flagship-tier buyer would be able to run.
Market Impact: NVIDIA’s Push Beyond Data Center GPUs
NVIDIA’s data center GPU business has dwarfed its gaming and consumer segments for several quarters running, and the company’s stock price has largely tracked that AI infrastructure demand rather than PC market conditions, a dynamic covered in depth around our reporting on NVIDIA’s roughly $500 billion AI infrastructure commitments. RTX Spark represents NVIDIA pushing the same Grace-and-Blackwell architecture it built for hyperscale data centers down into a product category it has to sell one unit at a time to individual developers and creative professionals, a far slower and more fragmented go-to-market than selling GPU clusters to cloud providers.
That shift lands at an awkward moment for GPU pricing more broadly. Our coverage of NVIDIA GPU prices spiking 41% in Japan and the broader RTX 5090 shortage pushing resale prices as high as 5x MSRP both point to the same underlying pressure: AI demand for Blackwell-generation silicon is straining supply across NVIDIA’s entire product stack, consumer GPUs included. Launching a new laptop superchip platform into that same supply environment raises real questions about how consistently NVIDIA and its six OEM partners can keep RTX Spark laptops in stock through the holiday shopping season.
There is a parallel pressure on system memory. DDR5 RAM prices climbing toward $450 for 32GB kits as AI demand eats into memory chip supply makes NVIDIA’s decision to equip RTX Spark laptops with up to 128GB of LPDDR5X unified memory a far more expensive proposition today than it would have been a year ago, and that cost almost certainly factors into the $2,599 to $4,399.99 pricing spread across the launch lineup.
How RTX Spark Fits Against Rival AI Laptop Platforms
RTX Spark does not arrive into an empty field. Intel, AMD, and Qualcomm have all spent the past two years marketing their own laptop chips around on-device AI capability, typically built around a dedicated neural processing unit sitting alongside a conventional CPU and integrated or discrete GPU. Apple’s unified memory architecture across its M-series chips has also become a reference point the industry measures local AI capability against, since Apple proved years ago that sharing memory between CPU and GPU cores, rather than partitioning it, unlocks real workloads that split-memory designs struggle with.
What makes RTX Spark different from the Intel, AMD, and Qualcomm approach is that NVIDIA is not adding an AI accelerator to an existing CPU design, it is building the entire chip, CPU and GPU together, around AI and graphics workloads from the start, with the full weight of its Blackwell GPU architecture rather than a scaled-down neural processing block. That is a much more direct architectural answer to Apple’s unified memory pitch than anything Intel, AMD, or Qualcomm has shipped in a Windows laptop to date. TechRadar frames the competitive dynamic bluntly, noting that even NVIDIA’s own far pricier Dell Pro Max 18 Plus mobile workstation, built around a different NVIDIA GPU tier, still holds the title of fastest AI laptop for now, suggesting RTX Spark is positioned as the accessible on-ramp to local AI computing rather than the absolute performance ceiling.
The PC chip landscape RTX Spark is entering has also been volatile on the stock side. Our look at Intel and AMD shares sliding while NVIDIA ticked up on PC chip competition shows investors are already pricing in NVIDIA encroaching further into territory Intel and AMD have long treated as their own, and a six-OEM laptop launch built on NVIDIA’s own CPU design is about as direct a challenge to that territory as the company has made yet.
Historical Context: From DGX Spark to Laptop Spark
The Spark name did not start with laptops. NVIDIA’s DGX Spark was a desktop-form-factor appliance the company positioned as a personal AI supercomputer, aimed at researchers and developers who needed serious local AI compute without the cost or complexity of a full data center rack. RTX Spark laptops extend that same underlying Grace-Blackwell superchip concept into a mobile form factor OEM partners can sell through ordinary consumer and enterprise retail channels rather than through NVIDIA’s more specialized DGX sales process.
That progression mirrors how NVIDIA has built out its broader AI model portfolio too. The company has been shipping a steady cadence of open AI models alongside its hardware pushes, including the recent Nemotron 3.5 family spanning three model sizes from 4GB to 120B parameters, giving developers a software stack tuned to run well on NVIDIA’s own silicon from day one. Pairing new hardware launches with ready-to-run open models has become a consistent NVIDIA playbook, and RTX Spark appears to be following the same script: sell the chip, then make sure there are models optimized to run on it the moment buyers unbox the laptop.
Competitive Comparison: RTX Spark vs. NVIDIA’s Own Pro Max Lineup
One of the more unusual competitive dynamics in this launch is that RTX Spark’s sharpest rival may be another NVIDIA-powered machine. TechRadar’s reporting on the Dell Pro Max 18 Plus, a separate mobile workstation line priced as high as $18,600 and built around a different, higher-tier NVIDIA GPU, frames that machine as still holding the fastest-AI-laptop crown even after RTX Spark’s launch. That positions RTX Spark less as NVIDIA’s top-of-the-line local AI hardware and more as the volume play meant to put Grace-Blackwell architecture into far more hands at a fraction of the Pro Max price.
That internal segmentation gives NVIDIA room to sell both ends of the market without the two product lines cannibalizing each other. Buyers who need absolute peak local AI performance and have enterprise budgets to match can still step up to Pro Max-class hardware. Buyers who want a meaningful taste of Grace-Blackwell’s unified memory advantage without a five-figure price tag now have RTX Spark starting at $2,599, a gap wide enough that NVIDIA’s OEM partners can market both tiers simultaneously without confusing buyers about which machine fits which budget.
What Early Buyers and Reviewers Are Flagging
Early coverage of the RTX Spark launch has focused heavily on the price-to-capability tradeoff rather than raw benchmark numbers, since independent third-party testing of retail units is still in its early days following the October 16 on-sale date. PCMag’s preorder coverage frames the machines as pricey but defensible given what they offer, a framing that lines up with NVIDIA’s own positioning of RTX Spark as a professional and developer tool rather than a mainstream consumer laptop competing with sub-$1,500 ultrabooks.
Regional retail coverage has also picked up the launch quickly. Outlets including Tbreak and GadgetPilipinas have reported on RTX Spark availability and the $2,599 starting price reaching international markets alongside the US launch, which suggests NVIDIA and its OEM partners coordinated a genuinely global rollout rather than a US-first staggered release, a departure from how some premium laptop launches have unfolded in the past.
The Arm-based Windows question remains the biggest open concern reviewers are likely to probe once more units reach hands-on testing. Windows on Arm has improved substantially in recent years, but application compatibility, particularly for specialized creative and engineering software that still assumes x86 architecture, remains a real variable that could shape how enthusiastically professional buyers adopt RTX Spark regardless of how compelling the unified memory story is on paper.
Predictions: Where RTX Spark Goes From Here
A handful of trends seem likely to play out over the coming months as RTX Spark laptops reach more buyers and reviewers.
- Expect Intel, AMD, and Qualcomm to accelerate their own unified-memory and on-device AI messaging in response, even if their underlying architectures do not fully match Grace-Blackwell’s shared memory design in the near term.
- Supply constraints already hitting NVIDIA’s RTX 5090 desktop GPU and broader DDR5 memory pricing make it plausible that RTX Spark laptop availability tightens heading into the holiday shopping season rather than loosening.
- More OEMs beyond the initial six are likely to announce RTX Spark models at CES in January, following NVIDIA’s typical pattern of expanding hardware partner rosters shortly after an initial launch window.
- Software compatibility tooling for Arm-based Windows will likely see renewed investment from Microsoft and NVIDIA alike, given how directly RTX Spark’s success depends on professional creative and development software running smoothly without emulation penalties.
- Pricing on the lower-tier 5,120-core, 32GB configuration will likely come under pressure to drop as memory supply stabilizes, since the $2,599 entry price currently sits close to the flagship configuration’s starting cost despite the significant memory and core count gap between the two tiers.
What This Means for AI PC Buyers Right Now
For most everyday laptop buyers, RTX Spark is not the machine to chase. The $2,599 starting price and the specific appeal of its architecture, local large language model inference and development, are aimed at a narrower professional and developer audience than the broader premium laptop market. Buyers who mainly want faster gaming performance or longer battery life in a thinner chassis are likely better served by conventional discrete-GPU laptops that do not carry the Arm-based Windows compatibility questions RTX Spark machines bring with them.
For developers and AI researchers who have been renting cloud GPU time just to prototype with larger open-weight models, RTX Spark is the first mainstream retail laptop platform built specifically to bring that workload onto a machine they can own outright. Whether the economics beat a well-managed cloud GPU budget will depend heavily on how often that compute gets used, but the option to own 128GB of unified memory in a portable chassis starting at $2,599 did not exist in the Windows laptop market before this week.
Frequently Asked Questions
What is the NVIDIA RTX Spark laptop?
It is a new laptop platform built around a superchip that combines a Blackwell RTX GPU with NVIDIA’s own Arm-based Grace CPU on a single package with unified memory, sold through ASUS, Dell, HP, Lenovo, Microsoft, and MSI starting October 16, 2026.
How much does an RTX Spark laptop cost?
Prices start at $2,599, with the Microsoft Surface Laptop Ultra and ASUS ProArt P14 both listed at $2,599.99. The larger ASUS ProArt P16 RTX Spark starts at $4,399.99.
What chip configurations are available?
Two configurations exist: a flagship RTX Spark N1X with a 6,144-core Blackwell GPU, a 20-core Grace CPU, and up to 128GB of unified LPDDR5X memory, and a lower-tier option with a 5,120-core GPU, an 18-core Grace CPU, and up to 32GB of unified memory.
Which operating system do RTX Spark laptops run?
Windows 11. The Grace CPU is Arm-based, which means RTX Spark laptops run on Windows on Arm rather than the traditional x86 architecture most Windows software was built for.
Why does unified memory matter for RTX Spark?
Unified memory lets the CPU and GPU share the same memory pool instead of splitting system RAM from dedicated video memory. On the flagship configuration that pool scales to 128GB, letting developers load larger AI models locally than conventional discrete-GPU laptops typically allow.
How much AI performance does RTX Spark offer?
NVIDIA lists up to 1 petaflop of AI performance for the flagship 6,144-core, 20-core configuration. NVIDIA has not published a separate AI performance figure for the lower-tier 5,120-core configuration.
Which laptop brands are selling RTX Spark hardware?
ASUS, Dell, HP, Lenovo, Microsoft, and MSI are confirmed launch partners, with HP and Microsoft already listing models through their own retail storefronts.
Is RTX Spark meant for gaming or for AI work?
NVIDIA is positioning RTX Spark primarily for developers, AI researchers, and creative professionals who want to run local AI workloads, rather than as a dedicated gaming laptop platform, though its Blackwell GPU and ray-tracing cores will handle games as well.
