Microsoft has put a price and a ship date on its most powerful Surface laptop ever. The Surface Laptop Ultra, built around Nvidia’s RTX Spark Superchip, starts preorders at $2,599 and begins shipping October 16, 2026. Microsoft is pitching it less as a laptop upgrade and more as a local AI workstation that happens to fold shut: the company says it can run AI models with up to 120 billion parameters without touching the cloud.
That claim is the headline here. Running a 120-billion-parameter model locally, on battery power, inside a laptop chassis, was not something consumer hardware could credibly promise a year ago. Microsoft says the Surface Laptop Ultra changes that math, and the company is betting $2,599 is a price people will pay to find out.
What Microsoft Actually Announced
Preorders for the Surface Laptop Ultra opened October 7, 2026, with Microsoft confirming retail availability starting October 16, a timeline also reported by The Verge and CNBC. The $2,599 MSRP covers the entry configuration; Microsoft has not published full base-model specifications for CPU core count, RAM, or storage, and early reporting has conflicted on those exact numbers, so treat any specific base-tier spec breakdown you see elsewhere with caution until Microsoft posts a full spec sheet.
What Microsoft has confirmed is the chip at the center of the machine: an NVIDIA RTX Spark Superchip, which pairs a Blackwell-generation RTX GPU with an Nvidia Grace CPU carrying up to 20 CPU cores. That’s a notable architecture choice. Instead of building around an in-house Arm chip or a conventional x86 part from Intel or AMD, Microsoft turned to Nvidia’s combined GPU-plus-CPU superchip design, the same family of silicon Nvidia has been pushing into desktop AI boxes like the DGX Spark and that other PC makers have started slotting into their own machines, including ASUS’s ProArt RTX Spark workstation.
Microsoft describes the Surface Laptop Ultra as “our most powerful Surface laptop ever, built for local AI and 3D graphics.” The company is leaning on two numbers to back that up: up to 128 GB of unified memory, and up to 1 petaflop of AI compute. Unified memory is the detail that matters most for the local-AI pitch, since it’s what lets a model’s weights and the GPU’s working memory share the same pool instead of being capped by a smaller, dedicated VRAM block.
The 120-Billion-Parameter Claim, In Context
Microsoft’s specific claim is that the Surface Laptop Ultra can run AI models with up to 120 billion parameters locally. To put that in perspective: most “run it on your laptop” open models that have circulated so far top out well below that, usually in the 7-billion to 70-billion parameter range, and even then often require quantization tricks to fit in consumer memory. A 120-billion-parameter model running without a cloud round-trip is a meaningfully different class of machine than a typical AI PC shipping today.
The pitch lines up with how Microsoft has been talking about AI on Windows all year. CEO Satya Nadella has repeatedly framed Microsoft’s strategy as blending cloud and on-device intelligence rather than picking one. Speaking about Microsoft’s approach around the Surface Laptop Ultra launch, Nadella said: “The trajectory for me is so clear. There will never be a moment where we will go and look and say: ‘Oh, this runs locally, this runs in the cloud.’ You will expect this hybrid intelligence to be everywhere and pervasive,” according to France 24.
Nadella also made the case that hardware alone doesn’t deliver the payoff Microsoft is promising. “Just having the [AI] model doesn’t do much for anything,” he said, according to Fortune, a remark made while explaining Microsoft’s effort to tie local models, AI-aware Windows software, and AI-native hardware like the Surface Laptop Ultra into one package rather than selling the chip on its own. He went further on the software layer, telling the same event that “Copilot now truly understands your PC,” per Fortune’s reporting on the launch.
That’s the actual sales pitch: not “buy a faster laptop,” but “buy the laptop that makes Windows Copilot and on-device AI features work the way Microsoft has been promising they eventually would.” Whether that justifies $2,599 is the question reviewers and buyers will spend the next few weeks answering.
Pricing and Availability Timeline
| Milestone | Date | Detail |
|---|---|---|
| Preorders open | October 7, 2026 | Surface Laptop Ultra listed starting at $2,599 MSRP |
| Retail shipping begins | October 16, 2026 | Confirmed by Microsoft EVP Pavan Davuluri |
| Chip platform | Announced alongside launch | NVIDIA RTX Spark Superchip (Blackwell RTX GPU + up to 20-core Grace CPU) |
| Max unified memory | Announced alongside launch | Up to 128 GB |
| Claimed AI compute | Announced alongside launch | Up to 1 petaflop |
| Local model ceiling | Announced alongside launch | Up to 120 billion parameters, per Microsoft |
Pavan Davuluri, Microsoft’s executive vice president of Windows and devices, is the executive who put a firm date on shipping, confirming the laptop would reach buyers on October 16. That’s a nine-day gap between preorder and delivery, which is tight by Surface standards and suggests Microsoft has supply already staged rather than announcing a device it expects to trickle out over months.
Why Microsoft Picked Nvidia Silicon Over Its Own Arm Chips
For several product cycles, Microsoft’s higher-end Surface machines leaned on Qualcomm’s Snapdragon X-series Arm chips for the “AI PC” story, including the Snapdragon X2-based refresh that started at $1,149 and claimed 80 TOPS of NPU performance. The Surface Laptop Ultra is a different animal entirely: instead of an NPU bolted onto a general-purpose Arm chip, it’s built around a full GPU-plus-CPU superchip designed from the ground up for AI workloads at the silicon level.
Nvidia’s own product pages for the Grace CPU architecture and the DGX Spark platform describe the same design goal Microsoft is now shipping inside a laptop: a CPU and GPU sharing one coherent memory pool so large model weights don’t have to be split, swapped, or quantized down just to fit. That’s a very different engineering bet than the NPU-first approach most “AI PC” marketing has used since 2024.
That’s also a notable shift because, as covered separately, Microsoft’s Surface line has gone roughly five years without shipping an AMD-powered configuration, instead rotating between Intel, Qualcomm, and now Nvidia-class silicon. Pairing with Nvidia puts Microsoft in the same camp as PC makers like ASUS, which has already shipped an RTX Spark-based ProArt workstation advertised around the same 128 GB memory ceiling and 120-billion-parameter local model claim, and HP, which briefly listed its own RTX Spark laptop at $2,999.99 before pulling the listing. Microsoft undercutting that HP price point by roughly $400 is likely not a coincidence. Tom’s Hardware notes the launch also coincides with broader Windows 11 changes Microsoft is rolling out for agentic AI features.
The Surface Laptop Ultra’s debut also follows closely on the heels of earlier reporting that pegged its specs at 128 GB of RAM and roughly 6,144 GPU cores well before Microsoft’s official pricing landed. That earlier leak lines up with what Microsoft has now confirmed on memory, lending the leak credibility in hindsight, even though Microsoft itself has not published a core-count figure as part of this announcement.
Competitive Landscape: Where the Surface Laptop Ultra Sits
The RTX Spark platform is quickly becoming a crowded field rather than an Nvidia-and-Microsoft exclusive. Nvidia has already sold its own DGX Spark desktop boxes in two memory tiers, and multiple OEMs have announced RTX Spark-based machines within weeks of each other. That makes the Surface Laptop Ultra one entrant in a fast-moving category rather than a singular product.
| Device | Maker | Price | Max Memory | Form Factor |
|---|---|---|---|---|
| Surface Laptop Ultra | Microsoft | $2,599 (starting) | Up to 128 GB | Laptop |
| RTX Spark laptop | HP | $2,999.99 (briefly listed) | Not confirmed | Laptop |
| ProArt RTX Spark | ASUS | Not yet confirmed publicly | Up to 128 GB | Desktop workstation |
| DGX Spark | Nvidia | $4,999 (64GB tier) | 128 GB (top tier at $6,950) | Desktop box |
Two things stand out in that lineup. First, Microsoft’s $2,599 starting price undercuts both HP’s briefly-listed RTX Spark laptop and Nvidia’s own 128 GB DGX Spark desktop tier, despite being a portable machine rather than a desk-bound box. Second, laptop form factor is itself the differentiator here: Nvidia’s DGX Spark line and ASUS’s ProArt machine are desktop-class products, while Microsoft and HP are racing to be first to put this silicon into something you can close and carry. If the 120-billion-parameter local inference claim holds up under independent testing, the Surface Laptop Ultra would be the cheapest way to get that capability in a laptop today.
Historical Context: How We Got Here
The “AI PC” label has been attached to Windows hardware since 2024, mostly tied to NPUs rated in the tens of TOPS, good for tasks like webcam background blur, live captioning, and small on-device inference features in Copilot. Those machines were never positioned to run anything close to a 120-billion-parameter model; their NPUs weren’t built for it, and most didn’t have the unified memory to hold a model that large even if the compute existed.
The jump to GPU-plus-CPU superchip designs, the approach Nvidia has pushed through its Grace-Blackwell combinations, changes that equation. Instead of an NPU handling light inference tasks alongside a separate CPU and GPU, the RTX Spark architecture puts a full Blackwell RTX GPU and a many-core Grace CPU on a shared memory pool. That’s architecturally closer to what data-center AI accelerators do than what a typical laptop chip does, scaled down into something that fits a laptop’s thermal envelope. Microsoft’s decision to build the Surface Laptop Ultra around that platform, rather than iterating on the NPU-based Snapdragon approach it used for the Surface Refresh earlier this year, signals the company sees on-device large-model inference, not just lightweight NPU acceleration, as the next phase of the “AI PC” pitch.
What’s Still Unconfirmed
Microsoft’s announcement leaves real gaps. The company has not published base-configuration specifications covering CPU core count, RAM capacity, or storage size for the entry $2,599 tier, and outlets covering the launch have reported conflicting numbers on those specifics. Until Microsoft posts a full spec sheet or third-party reviewers get hands-on units after the October 16 ship date, buyers should treat any specific core-count or storage figure circulating online as unverified.
It’s also worth noting that “AI PC” in coverage of this launch is a description, not Microsoft’s official product name; the device is marketed as the Surface Laptop Ultra. And the 120-billion-parameter local inference figure is Microsoft’s own claim. Independent benchmarking after the October 16 ship date will determine how that claim holds up across different model architectures and quantization levels, something Microsoft hasn’t detailed.
Who Is the Surface Laptop Ultra Actually For?
Microsoft’s marketing language, “built for local AI and 3D graphics,” points at two distinct buyers rather than one. The 3D graphics half targets the same creative professionals who’d normally look at a mobile workstation: people doing 3D rendering, video work, or CAD who want GPU headroom without shipping their files to a render farm. The local AI half targets developers, researchers, and AI engineers who want to run and fine-tune open-weight models without racking up cloud GPU bills or sending proprietary data to a third-party API.
Those two audiences overlap less than Microsoft’s pitch might suggest, which is part of why the Surface Laptop Ultra reads as a niche flagship rather than a mainstream refresh. It sits above the Snapdragon-based Surface lineup in price and capability, aimed at the slice of Surface’s audience that outgrows integrated NPU acceleration and needs real local GPU compute instead.
Market Impact: What This Means for the PC Industry
A $2,599 laptop is not a mass-market product, and Microsoft isn’t pretending otherwise. The more important signal is what the Surface Laptop Ultra tells other PC makers about where the premium end of the Windows lineup is headed. If Microsoft, Nvidia’s own hardware partner for Windows AI features, is willing to put RTX Spark silicon in a flagship Surface rather than sticking with Qualcomm or Intel at the high end, that’s a strong nudge for Dell, Lenovo, and others to follow with their own Grace-Blackwell laptops.
It also puts pressure on Apple’s side of the market. Apple has been selling its own on-device AI story around unified memory in M-series chips, most recently with the M5 Ultra’s 36-core design and 1.2TB/s of memory bandwidth. Microsoft’s 128 GB unified memory figure and 120-billion-parameter claim are a direct answer to that pitch, aimed at developers and AI-heavy professionals who might otherwise default to a Mac for local model work.
For Nvidia, the Surface Laptop Ultra is validation that RTX Spark isn’t just a desktop or workstation story. Landing inside a Surface, Microsoft’s own flagship hardware line, gives the chip platform credibility it wouldn’t get from smaller OEM design wins alone. Expect Nvidia to point to this launch repeatedly in its own marketing over the next few quarters.
Predictions: Where This Goes Next
- Expect at least two more OEMs (likely Dell and Lenovo, given their history of shadowing Surface hardware launches) to announce RTX Spark laptops within the next two to three months.
- Independent reviewers will almost certainly find real-world local-model performance falls short of the “120 billion parameters” headline figure for some model families, particularly ones not optimized or quantized for the Grace-Blackwell memory architecture.
- Microsoft will likely use the Surface Laptop Ultra as the reference hardware for new Copilot features that specifically require local inference, framing those features as “best on Surface Laptop Ultra” the way it has done with NPU-gated Copilot+ features in the past.
- Pricing on competing RTX Spark laptops from HP and others will probably settle closer to Microsoft’s $2,599 than the $2,999.99 HP briefly listed, as competitors react to Microsoft’s number.
- Expect Microsoft to publish full base-configuration specs (CPU cores, RAM, storage) before or shortly after the October 16 ship date, closing the gap left by this announcement.
Should You Preorder the Surface Laptop Ultra?
For most Windows users, no. $2,599 is a lot of money for a laptop whose main differentiator, running very large AI models locally, is a niche capability outside of developers, researchers, and AI-focused professionals. A buyer who just wants faster Copilot features and longer battery life is better served by a Snapdragon X2 Surface Refresh at under half the price.
For developers and technical teams who specifically want to run large open-weight models without cloud API costs or data leaving the device, the Surface Laptop Ultra is one of the first portable machines that makes that plausible at all. Whether it’s worth the price over a desktop RTX Spark box, or over Nvidia’s own DGX Spark, comes down to whether portability matters enough to pay a premium and accept whatever thermal throttling a laptop chassis imposes on sustained AI workloads compared to a desktop box.
Frequently Asked Questions
How much does the Surface Laptop Ultra cost?
The Surface Laptop Ultra starts at $2,599 MSRP for its entry configuration. Microsoft has not published pricing for higher-memory configurations beyond confirming the laptop supports up to 128 GB of unified memory.
When does the Surface Laptop Ultra ship?
Preorders opened October 7, 2026, and Microsoft EVP Pavan Davuluri confirmed retail shipping begins October 16, 2026.
What chip powers the Surface Laptop Ultra?
It uses the NVIDIA RTX Spark Superchip, which combines a Blackwell-generation RTX GPU with an Nvidia Grace CPU carrying up to 20 CPU cores.
Can the Surface Laptop Ultra really run AI models with 120 billion parameters locally?
That’s Microsoft’s own claim, based on the device’s up-to-128GB unified memory and up-to-1-petaflop AI compute. Independent testing after the October 16 ship date will show how that claim holds across different model types and quantization settings.
How does the Surface Laptop Ultra compare to Nvidia’s DGX Spark?
Both use RTX Spark-class Grace-Blackwell silicon and top out at 128 GB of unified memory, but the DGX Spark is a desktop box starting at $4,999, while the Surface Laptop Ultra is a portable laptop starting at $2,599.
Is “Surface Laptop Ultra” an official Microsoft product name?
Yes. “AI PC” used in headlines about the launch is a descriptive term from coverage, not an official Microsoft product name; Microsoft calls the device the Surface Laptop Ultra.
What are the base-model specs for CPU, RAM, and storage?
Microsoft has not published those figures for the entry $2,599 configuration, and early reports conflict on the exact base-tier specification. Treat specific base-tier numbers you see elsewhere as unconfirmed until Microsoft releases a full spec sheet.
Does the Surface Laptop Ultra replace Microsoft’s Snapdragon-based Surface line?
No. The Snapdragon X2-based Surface Refresh, starting around $1,149, remains Microsoft’s mainstream AI PC option. The Surface Laptop Ultra is a separate, higher-end line aimed at local large-model AI work and 3D graphics.
