Microsoft Surface Laptop Ultra Ships with NVIDIA RTX Spark Superchip and Up to 128 GB Unified Memory

Introduction

Microsoft opened preorders on October 7, 2026 for the Surface Laptop Ultra and the Surface RTX Spark Dev Box, the first two devices built entirely around NVIDIA's RTX Spark Superchip. The announcement, made through the Windows Blog, closes a loop that began in the spring when NVIDIA and Microsoft jointly announced RTX Spark as the foundation for a new class of Windows on Arm systems. What made that spring announcement theoretical — a silicon platform with no shipping hardware — becomes concrete on October 16, when the laptop goes on sale at a starting price of $2,599.

The hardware answer Microsoft landed on is unusual enough to be worth stating plainly. The RTX Spark Superchip is not a conventional laptop processor with an integrated GPU bolted on. It is a single package that combines an NVIDIA Blackwell RTX GPU with up to 6,144 cores, an NVIDIA Grace CPU with up to 20 Arm cores, and — the number that matters most — up to 128 GB of unified LPDDR5X memory shared dynamically between the CPU and the GPU. Microsoft claims one petaflop of AI performance, enough, in its framing, to run AI models exceeding 120 billion parameters locally on a laptop that is under 18 mm thin and under 4.5 pounds.

For the semiconductor industry, the significance is less about Microsoft's laptop business and more about where this leaves competition at the top of the PC stack. Apple's MacBook Pro remains the reference point for this class of machine, and Microsoft's specification sheet is explicitly aimed at it: the Surface Laptop Ultra delivers, by Microsoft's own measurement, up to 25 percent brighter peak HDR than a MacBook Pro M5 Pro, in a chassis that competes at the same weight. The question is whether an Arm-based Blackwell part can finally close the performance and software gaps that have dogged Windows on Arm for a decade. Full context on the category is available at semiconductors.

The Silicon: What RTX Spark Actually Is

A Superchip Born in the DGX Spark

The RTX Spark Superchip did not begin life as a laptop part. NVIDIA introduced its underlying technology as the Grace Blackwell GB10 Superchip, the processor at the heart of the DGX Spark desktop system. The laptop variant compresses that architecture into a mobile power envelope while retaining the unifying feature: CPU and GPU share one memory pool. NVIDIA's RTX Spark product page lists the headline specification as 6,144 Blackwell RTX cores, a 20-core Grace CPU, one petaflop of FP4 AI performance and 128 GB of unified memory.

That unified memory design is the structural departure from how almost every existing laptop works. A conventional x86 laptop with a discrete NVIDIA GPU carries two physically separate memory pools — system DRAM and video RAM — connected by a bus that forces the CPU to copy data across before the GPU can use it. RTX Spark eliminates that copy. A large language model loaded into unified memory is equally accessible to the CPU and the GPU without a transfer step, and the CPU's workload and the GPU's workload can contend for the same gigabytes without either hitting a wall.

MediaTek's Role in the Physical Package

The SoC is co-designed and manufactured with MediaTek, a partnership that both companies confirmed in 2026 and that is reflected in the part's reported internal designation. TechPowerUp's database catalogues the chip as the MediaTek AHJ11488B with a thermal design power of 38 watts, while its GPU-side entry describes a 128 GB LPDDR5X memory subsystem on a 256-bit interface. The MediaTek product page presents the identical specification triad of 6,144 GPU cores, 20 CPU cores and one petaflop of FP4 performance.

That arrangement is a meaningful signal about how advanced SoCs now get built. NVIDIA owns the GPU architecture, the Grace CPU design and the software stack, but the physical integration — die placement, power delivery, memory interfaces, package engineering — is shared with a partner that has decades of mobile SoC shipping experience. The result is that RTX Spark's silicon genealogy is less like a desktop CPU and more like a smartphone processor that happens to contain a full Blackwell GPU.

The Hardware: Two Devices, One Platform

Surface Laptop Ultra

The Surface Laptop Ultra is the flagship. Its specification ladder is gated on the silicon you choose. The entry configuration pairs a 5,120-core Blackwell RTX GPU with an 18-core Grace CPU and 24 GB of unified memory, with 512 GB of storage, starting at $2,599. Higher tiers step up to the 6,144-core, 20-core part with 32 GB, 48 GB, 64 GB or 128 GB of unified memory, according to Microsoft's configuration page and independent notebookcheck coverage.

The display is a 15-inch PixelSense Ultra touchscreen that Microsoft claims is the brightest and sharpest ever put on a Surface Laptop, with a peak HDR brightness of 2,000 nits measured over a 10 percent HDR window. Microsoft's footnotes are unusually specific about how that measurement is made: peak HDR only, a 10 percent window, sustained and SDR brightness lower, conditions varying. The comparison the company chose to publish is that this is up to 25 percent brighter than the MacBook Pro M5 Pro's published peak HDR specification.

Thermal design was a stated constraint. Microsoft says the new thermal architecture delivers up to 2.5 times more thermal power capacity than its current Surface Laptops, in a chassis under 18 mm thick and under 4.5 pounds. Those are the two numbers that determine whether a 128 GB unified memory system can sustain high compute load, and they are the least glamorous part of the specification and the most operationally decisive.

The port selection includes something genuinely novel: Magnetic Connect, which Microsoft describes as the world's first built-in magnetic USB-C charging on a laptop. The included cable attaches magnetically to the right-side USB-C port and releases easily, and the port remains fully functional for video, data and other USB-C devices. Beyond that the laptop carries two additional USB-C ports, HDMI, USB-A, an SD card reader and a 3.5 mm headphone jack, and Microsoft claims support for up to three 4K external displays. Storage is user-removable.

Surface RTX Spark Dev Box

The second device drops the display entirely. The Surface RTX Spark Dev Box is a compact small form factor desktop built around the same RTX Spark Superchip, priced at $5,999 and available exclusively on Microsoft.com in the U.S. with shipping beginning in November. It is a member of what Microsoft calls the Project Zenith family and ships preconfigured with Visual Studio Code, Git, the GitHub CLI, GitHub Copilot, WSL, Python and Node.

The pitch to developers is explicit: move AI experiments off cloud infrastructure. Microsoft frames the value as iteration speed and cost control rather than raw throughput — run inference, evaluation and prototyping locally, reserve cloud capacity for the workloads that genuinely need it. That framing competes directly with the cloud GPU rental market on a cost-per-experiment basis, which is a different economic argument than the laptop's.

The Software Angle: Agents, Containers and Governance

The silicon announcement is wrapped in a software story that Microsoft is arguably more invested in. Microsoft Execution Containers provide containment for AI agents running on the device, and upcoming Windows capabilities will use Microsoft Entra to distinguish agent activity from user activity while extending Agent 365 controls to local on-device agents. The combination is meant to answer the question that has made enterprise IT hesitant about local AI at scale: if a model and its agent run on the laptop, who governs the access, the compute and the cost?

Microsoft's answer is a hybrid claim: Surface is the hardware for Windows, Windows is the operating system for agents, and the cloud remains available through Microsoft Dev Box when local capacity runs out. Whether the governance layer actually delivers on that promise is unverified by any independent source at announcement time, and the distinction matters for organisations evaluating the platform on compliance grounds.

The Competitive Frame: Taking On Apple

The most consequential comparison in the announcement is with Apple, and Microsoft did not leave it to interpretation. Alongside the HDR brightness claim, the company is running a trade-in promotion offering up to $1,000 cash back when trading in an eligible MacBook Pro, valid for purchases between October 7 and November 23, 2026. Independent coverage has been consistent in reading the positioning this way. The Verge's roundup treats the Surface Laptop Ultra as the biggest announcement of the event, Mashable and Tom's Guide both describe it as a direct challenge to the MacBook Pro, and ServeTheHome confirmed the shipping lineup for both devices.

Apple's counterargument is the one that historically decides this class: not peak silicon specification but the installed base of native applications. Windows on Arm has improved substantially through emulation and an expanding native catalogue, and Microsoft points to CUDA compatibility and what it calls the full Windows AI stack plus ray tracing, DLSS and NVIDIA Reflex in supported titles as the answer. Whether Blackwell's Arm-based CPU closes the last of the software gap is the question that the October 16 availability date begins to answer, not one that a press release can settle.

What This Means for the Chip Industry

Two structural points follow from the announcement. The first is that a supercomputer-class memory design is now shipping in a consumer laptop at a four-figure price. The DGX Spark established 128 GB of unified memory as a desktop workstation feature; RTX Spark carrying the same configuration in a mobile chassis is a compression of an entire product tier into a form factor that fits in a bag.

The second is that advanced SoC development has clearly entered a co-design era. NVIDIA supplies the GPU and CPU architecture and the CUDA ecosystem, MediaTek supplies the package and integration expertise, and Microsoft supplies the device engineering and the operating system. Three companies' specialisations collapse into one SKU. That is a different industrial model from the era when a chipmaker designed the die, fabbed the die, shipped the die, and called it a product.

Conclusion

The Surface Laptop Ultra and the Surface RTX Spark Dev Box are the first shipping hardware built around NVIDIA's RTX Spark Superchip, and they put a specification that was previously confined to a desk-side AI workstation into laptops starting at $2,599. The unified memory architecture is the substantive change, not the core counts, and it is the part that determines whether running 120-billion-parameter models locally is a benchmark claim or a working tool.

The October 16 availability date and the $5,999 Dev Box's November shipping start make this an immediately testable proposition rather than a roadmap slide. Microsoft has chosen the most demanding possible comparison — Apple's MacBook Pro, on the same weight class, at comparable prices — and the next several months of independent benchmarking will determine whether the unified memory advantage is enough to convert into the sales and developer adoption the platform needs.

Images

References

• Microsoft Windows Blog: Pre-order our most powerful Surface devices ever
• NVIDIA: RTX Spark product page
• MediaTek: NVIDIA RTX Spark Superchip
• Notebookcheck: Microsoft Surface Laptop Ultra debuts with RTX Spark, 128 GB unified memory and $2,599 starting price
• ServeTheHome: Microsoft Launches Surface Laptop Ultra, Surface RTX Spark Dev Box
• The Verge: Everything announced at Microsoft's Surface Laptop Ultra event
• TechPowerUp: NVIDIA RTX Spark specifications
• Microsoft Store: Surface Laptop Ultra configurations