Microsoft is preparing to launch the Surface Laptop Ultra, a 15-inch Windows machine built around NVIDIA’s RTX Spark platform, with configurations topping out at 128GB of unified LPDDR5X memory and a 20-core Arm CPU paired with a 6,144-core Blackwell GPU. Notebookcheck published detailed specifications on October 6, 2026, while the Hungarian outlet news.lavx.hu had tracked the device since mid-June and updated its report in September. Preorders are expected to open October 7, 2026, with first deliveries arriving in mid-October, according to Notebookcheck’s reporting.
The timing matters. Microsoft is effectively bringing NVIDIA’s data-center-grade GB10 Grace Blackwell Superchip, the same silicon that powers the DGX Spark desktop AI system, into a laptop chassis for the first time. That’s a bigger deal than a routine spec bump. It puts a machine capable of running 120-billion-parameter models locally into the same product category as a 15-inch ultrabook, at a moment when memory prices are climbing and the AI PC category is still looking for its breakout device.
What Notebookcheck and news.lavx.hu actually reported
According to Notebookcheck’s October 6 write-up, the Surface Laptop Ultra pairs a Mini-LED display, rated up to 2,000 nits of peak HDR brightness, with NVIDIA’s RTX Spark chip in two tiers. The higher-end RTX Spark S3 configuration is reported to carry a 20-core Arm CPU and a 6,144-core Blackwell GPU, with memory options spanning 32GB, 48GB, 64GB, and 128GB. A lower S2 tier reportedly drops to an 18-core CPU and a 5,120-core GPU, with 24GB and 32GB memory choices.
News.lavx.hu, which first flagged the device back in June and refreshed its report on September 13, describes the 6,144-core Blackwell GPU as landing in the same graphics-performance neighborhood as an RTX 5070, while stressing that CUDA compatibility is the real selling point for developers who want to run AI workloads without converting code away from NVIDIA’s software stack. Windows Latest, reporting separately on October 3, pegged the platform’s AI throughput at roughly 1 petaflop and said it’s capable of running local models with up to 120 billion parameters, though those are platform-level claims from the outlet rather than independently verified benchmark results.
None of the three outlets reported a confirmed Microsoft price at publication time. News.lavx.hu’s own estimate puts a base configuration, reportedly 16GB of memory with a 512GB SSD, around $3,000, scaling up to roughly $7,000 for the fully loaded 128GB/4TB build. Those numbers are the outlet’s inference based on component cost and market positioning, not a confirmed Microsoft price list, and should be read with that caveat attached.
RTX Spark: from desktop AI box to laptop silicon
RTX Spark isn’t a new name on tech-insider.org. We’ve covered its debut as a Windows-on-Arm desktop chip arriving at $2,899 with 128GB of RAM, and its lineage traces directly back to the GB10 Grace Blackwell Superchip that NVIDIA built for the DGX Spark mini PC. NVIDIA’s official DGX Spark product page lists the chip’s Arm CPU complex as a mix of Cortex-X925 and Cortex-A725 cores, paired with 128GB of memory and up to 1 petaflop of AI compute in a box designed to sit on a desk next to a workstation.
Putting that same silicon class into a 15-inch laptop is a different engineering problem. Desktop and mini-PC chassis have room for active cooling solutions that a laptop simply doesn’t. If Microsoft’s Surface Laptop Ultra can sustain anything close to DGX Spark-level throughput inside a notebook body, thermal management, not raw compute, becomes the story worth watching once independent reviewers get hands-on units after the reported mid-October delivery window.
It’s also worth noting how fast NVIDIA’s own DGX Spark pricing has moved. The desktop unit launched at $3,999 and climbed to $4,699 by February 2026, a jump attributed to LPDDR5X supply constraints. By October 2026, NVIDIA’s own marketplace listing showed the DGX Spark priced at $6,950, with retail pricing varying by configuration and seller. That’s an 18% increase on the base model within months, driven by memory, not compute. A laptop sharing the same memory architecture inherits the same exposure.
Why 128GB of unified memory matters for a laptop
Unified memory architectures let the CPU and GPU draw from the same pool instead of copying data back and forth between separate VRAM and system RAM. For AI workloads, that’s the difference between a model fitting on-device or needing to be split, quantized aggressively, or offloaded to the cloud. A 128GB pool is large enough to hold a 70-billion-parameter model at higher precision, or a 120-billion-parameter model at reduced precision, entirely on a single machine without a network round-trip.
That’s the pitch behind the Surface Laptop Ultra’s top configuration, and it’s the same pitch NVIDIA has been making for DGX Spark since it started shipping in October 2025, following its announcement at GTC in March 2025. The difference this time is form factor: a laptop that can plausibly run a local AI assistant with no internet connection, no API bill, and no data leaving the device, which is a meaningfully different value proposition for developers, researchers, and enterprises with data-residency requirements than a desktop box tucked under a desk.
How it stacks up against Apple’s 128GB MacBook Pro
Apple has already staked out the high-memory professional laptop segment with the 2026 MacBook Pro line built on M5 Pro and M5 Max. The M5 Pro tops out at 64GB of unified memory, while the M5 Max scales to 128GB with up to 614GB/s of memory bandwidth. Apple’s own test configuration for the M5 Max pairs an 18-core CPU, a 40-core GPU, and that same 128GB ceiling.
Pricing gives a sense of where a 128GB Surface Laptop Ultra would likely land if news.lavx.hu’s estimate holds. A 16-inch MacBook Pro configured with an M5 Max chip, 128GB of memory, and maximum storage reportedly reached approximately $7,349 at launch, while a fully loaded 16-inch build with 8TB of storage and a nano-texture display was reported at $10,149. If Microsoft’s top Surface Laptop Ultra configuration does land near $7,000 as news.lavx.hu projects, that places it squarely inside Apple’s own pricing band for comparable memory capacity, rather than undercutting it.
We’ve tracked Apple’s side of this race closely, including the M6 chip’s 2nm leap over M5 and the MacBook Pro’s pending OLED redesign. The contest between Redmond and Cupertino for the high-memory professional laptop buyer is no longer theoretical; it’s a 2026 product category with real SKUs on both sides.
| Spec | Surface Laptop Ultra (RTX Spark S3, reported) | MacBook Pro 16″ (M5 Max) | NVIDIA DGX Spark (desktop) |
|---|---|---|---|
| CPU cores | 20-core Arm | 18-core | 20-core Arm (10x Cortex-X925 + 10x Cortex-A725) |
| GPU | 6,144-core Blackwell (~RTX 5070-class, per news.lavx.hu) | 40-core GPU | Blackwell-based GB10 GPU |
| Max unified memory | 128GB LPDDR5X | 128GB, up to 614GB/s bandwidth | 128GB |
| Form factor | 15″ Mini-LED laptop, up to 2,000 nits | 16″ laptop | Desktop mini PC |
| Reported/estimated top price | ~$7,000 (news.lavx.hu estimate) | $7,349–$10,149 (reported configurations) | $6,950 (NVIDIA marketplace listing, Oct. 2026) |
| Reported availability | Preorders Oct. 7, 2026; deliveries mid-October | Shipping (2026 lineup) | Shipping since Oct. 2025 |
The Surface family’s split personality
The Surface Laptop Ultra lands a few months after Microsoft brought Qualcomm’s Snapdragon X2 Plus chip to the Surface Pro at $1,149, and that contrast says a lot about where Microsoft sees its hardware line heading. The Snapdragon-powered Surface Pro is pitched at everyday productivity and battery life. The RTX Spark-powered Surface Laptop Ultra is pitched at developers and AI practitioners who want workstation-class local compute without buying a separate desktop tower.
That’s a notable shift in strategy for a product line that, as we’ve reported, went roughly five years without an AMD-powered Surface while leaning on Intel and then Qualcomm silicon. Bringing in NVIDIA’s Arm-based GB10 derivative as a third chip supplier for a flagship Surface model, rather than Intel, AMD, or Qualcomm, is itself a signal that Microsoft is treating AI compute density as a distinct product tier, separate from the general-purpose Copilot+ PC push it has run since 2024.
Memory prices are the wildcard
Every one of these 128GB machines, Surface Laptop Ultra included, depends on LPDDR5X supply that has gotten tighter and more expensive through 2026. We’ve covered how Micron’s CEO said memory supply is already 75% sold out for 2027, and how DRAM costs have spiked 4.4x in categories far removed from premium laptops, which tells you the pressure is systemic rather than isolated to one product line. Micron’s own fiscal 2026 results, a record $133 billion in revenue reported on Micron’s corporate site, underline how much of this capacity is already being absorbed by AI infrastructure buyers before it ever reaches a laptop bill of materials.
That’s the backdrop against which NVIDIA’s own DGX Spark went from $3,999 to $4,699 in roughly four months, an 18% increase tied directly to LPDDR5X supply, before climbing further to the $6,950 marketplace listing seen in October. A 128GB laptop variant carries the same soldered, large-capacity memory package, and the same exposure to that pricing trend. If Microsoft’s final price list comes in meaningfully above news.lavx.hu’s roughly $7,000 estimate for the top configuration, memory cost is the most likely explanation, not NVIDIA’s chip margins or Microsoft’s own markup.
Historical context: Surface’s long search for a developer machine
Microsoft has never quite had a Surface model that developers reach for by default, the way many reach for a MacBook Pro. The Surface Book line tried and was discontinued. The Surface Laptop Studio aimed at creatives more than coders. The standard Surface Laptop has always skewed toward business productivity rather than heavy local compute. The Surface Laptop Ultra, if the reported specs hold, is the first Surface model explicitly built around a workstation-class AI accelerator rather than an integrated GPU, which makes it the clearest attempt yet at a genuine high-performance developer machine under the Surface name.
That history matters for how seriously the market will take this launch. Enthusiasts and developers who passed on earlier Surface hardware because it wasn’t built for their workloads are exactly the audience Microsoft needs to convert now, and NVIDIA’s RTX Spark branding, already familiar from the desktop launch we covered earlier this year, gives Microsoft a recognizable performance story to lean on rather than starting from zero with a brand-new chip name.
Market impact: what this means for the AI PC category
The AI PC category has spent most of 2025 and 2026 defined by NPUs doing modest, on-device inference for things like live captioning and background blur. A 128GB laptop capable of running a 120-billion-parameter model locally, even if that figure is an outlet-reported platform claim rather than an independently benchmarked result, is a different tier entirely. It reframes “AI PC” from a marketing checkbox into a machine that can replace a cloud API subscription for certain workloads, which has real implications for software vendors pricing per-token access to large models.
It also puts pressure on competing Windows-on-Arm vendors. If the Surface Laptop Ultra’s reported specs and pricing hold up, it becomes the reference design other OEMs building on NVIDIA’s RTX Spark platform have to match or beat, the same way the original RTX Spark desktop launch set a $2,899 anchor price that subsequent OEM machines were measured against.
Competitive comparison beyond Apple
Apple isn’t the only comparison point. We’ve already benchmarked how the DGX Spark compares against Framework Desktop and Mac Studio, and that comparison is useful context here because it establishes how much of a gap exists between desktop-class unified-memory machines and what a laptop chassis can realistically sustain under thermal constraints. A Framework Desktop or Mac Studio can run fans at full tilt indefinitely; a 15-inch laptop cannot, which means sustained AI inference performance on the Surface Laptop Ultra is likely to trail its desktop-class siblings even with identical memory capacity.
| Product category | Example | Max unified memory | Form factor advantage |
|---|---|---|---|
| Laptop | Surface Laptop Ultra (reported) | 128GB | Portability, battery |
| Laptop | MacBook Pro 16″ M5 Max | 128GB | macOS ecosystem, battery life |
| Mini PC | NVIDIA DGX Spark | 128GB | Sustained thermal headroom |
| Desktop | Framework Desktop | Up to 128GB (platform-dependent) | Upgradability |
| Desktop/AIO | Mac Studio | Up to 512GB (higher-end configs) | Highest memory ceiling |
What’s still unconfirmed
It’s worth being precise about what’s reported versus what’s confirmed. Notebookcheck’s October 6 report and news.lavx.hu’s updated coverage describe specifications and a preorder timeline, but neither outlet, nor Windows Latest’s October 3 report, cites an official Microsoft press release with final pricing. The $3,000-to-$7,000 price range circulating is news.lavx.hu’s own estimate, built from comparable component costs, not a number Microsoft has published. Buyers and developers planning around this machine should treat the memory configurations and core counts as the most solid data points, since multiple outlets converge on them, and treat final pricing as the figure most likely to shift between now and an official Microsoft announcement.
Predictions: where this goes next
- Microsoft will likely confirm official pricing close to or at the reported October 7, 2026 preorder date, and the top 128GB configuration will probably land between $6,500 and $7,500 given how closely it tracks Apple’s M5 Max pricing band.
- Independent reviewers will focus heavily on sustained thermal performance once units ship in mid-October, since a laptop chassis running GB10-derived silicon at anything close to DGX Spark’s 1-petaflop claim is the single biggest unverified assumption in current reporting.
- Expect at least one other major Windows OEM to announce a competing RTX Spark laptop within two to three months, mirroring how quickly ASUS and MSI brought RTX Spark desktop machines to market after NVIDIA’s initial launch.
- Memory pricing will keep pressuring the entry point of these machines; if Micron’s 2027 sold-out supply signal holds, expect the lower 32GB and 48GB Surface Laptop Ultra configurations to see more aggressive retail promotion than the 128GB flagship, which will stay scarce and expensive.
- Apple is unlikely to let this launch pass without a response; a refreshed MacBook Pro marketing push emphasizing on-device AI model size, rather than raw GPU benchmarks, is a plausible counter-move given how directly the Surface Laptop Ultra targets the same 128GB unified-memory niche Apple has owned since the M5 Max launch.
Should developers and AI teams wait for this?
For teams already running local inference on a DGX Spark or a Mac Studio, the Surface Laptop Ultra doesn’t change the calculus much; it’s the same memory ceiling in a more portable, likely more thermally constrained package. For developers who need to move between locations and still want to prototype against a 70-to-120-billion-parameter model without a cloud bill, this is the first Windows laptop that makes that remotely realistic, assuming the reported specs survive contact with an official Microsoft announcement and independent benchmarking.
The more conservative move for most buyers is to wait for Microsoft’s confirmed pricing and the first round of independent reviews after the reported mid-October delivery window, rather than preordering against estimated figures. Given how much DGX Spark’s own price moved within months of launch, a wait-and-see approach on the Surface Laptop Ultra’s top configuration carries limited downside and avoids locking into a price point that reporting, not Microsoft, originally set.
Frequently asked questions
What is the Surface Laptop Ultra?
It’s a reported 15-inch Windows laptop from Microsoft built around NVIDIA’s RTX Spark chip platform, with configurations described by Notebookcheck and news.lavx.hu reaching 128GB of unified LPDDR5X memory and a 20-core Arm CPU with a 6,144-core Blackwell GPU.
When does the Surface Laptop Ultra launch?
Notebookcheck reports preorders opening October 7, 2026, with first deliveries beginning in mid-October 2026. Microsoft had not issued an official confirmation at the time of these reports.
How much does the Surface Laptop Ultra cost?
Official pricing has not been published. News.lavx.hu’s own estimate puts a base configuration around $3,000 and a fully loaded 128GB/4TB build around $7,000, but those are the outlet’s inferred figures, not confirmed Microsoft prices.
What is RTX Spark and how is it related to DGX Spark?
RTX Spark is NVIDIA’s branding for Windows-on-Arm devices built on the GB10 Grace Blackwell Superchip, the same chip family used in the DGX Spark desktop AI system. DGX Spark pairs a 20-core Arm CPU with 128GB of memory and up to 1 petaflop of AI compute, and the Surface Laptop Ultra’s reported top configuration mirrors those core specs in a laptop chassis.
How does it compare to the 128GB MacBook Pro M5 Max?
Both top out at 128GB of unified memory. The MacBook Pro M5 Max pairs that memory with an 18-core CPU, a 40-core GPU, and up to 614GB/s of memory bandwidth, with reported pricing between roughly $7,349 and $10,149 depending on configuration. If the Surface Laptop Ultra’s estimated ~$7,000 top price holds, the two machines would sit in a similar price band for comparable memory capacity.
Can the Surface Laptop Ultra run large AI models locally?
Windows Latest reports the platform is positioned at roughly 1 petaflop of AI compute and capable of running models with up to 120 billion parameters locally. These are outlet-reported platform claims rather than independently verified benchmark results, so real-world performance should be confirmed once review units are available.
Why are 128GB laptops getting more expensive across the industry?
LPDDR5X memory supply has tightened through 2026, with Micron’s leadership describing 2027 supply as already 75% sold out. NVIDIA’s own DGX Spark rose from $3,999 to $4,699 within months, an increase attributed to LPDDR5X constraints, and later marketplace listings showed pricing near $6,950. Any 128GB laptop, Surface Laptop Ultra included, carries similar memory-cost exposure.