Microsoft's Surface Laptop Ultra Breaks Qualcomm's Hold on Windows ARM with NVIDIA's RTX Spark

Among the Surface models Microsoft unveiled in October, one stands out for what is inside it. The Surface Laptop Ultra runs Windows on the ARM64 architecture, but it does not use a Qualcomm Snapdragon chip. It uses NVIDIA's RTX Spark, a system-on-chip that has no precedent in Windows ARM hardware.

Until now, ARM chips for Windows were effectively a Snapdragon-only affair. RTX Spark combines a Blackwell GPU with Cortex-X925 and A725 cores, up to 10 cores in total. It can be configured with up to 128GB of memory, a design that anticipates running large language models locally. Microsoft will not be the only maker to use it. Other OEMs are slated to ship RTX Spark machines, and an ASUSTeK Computer notebook has already come to light.

There was a technical logic behind the single-supplier arrangement. Limiting Windows ARM to one SoC family makes driver development easier, shortens the time needed to port Windows, and lightens the support load on PC makers. The source analysis argues that this does not fully explain it. In its reading, some kind of business relationship existed between Microsoft and Qualcomm, and as a result the porting target for Windows 10 and 11 stayed fixed on Snapdragon for a long time.

That arrangement now looks as if it is starting to crumble. One interpretation is that Qualcomm has been able to sell to parties other than Microsoft, and Snapdragon has indeed been adopted for Googlebook. Smartphones, tablets, and non-Windows PCs together far outsell Windows PCs, and most of them use ARM. Technical barriers to building hardware are low, which makes the architecture an attractive home for general-purpose AI inference chips.

The move also reshapes Microsoft's own lineup. Last year the company rolled out Copilot+ PC, which uses high-performance NPUs for local inference and offline AI use, but only with small-to-mid-sized language models. RTX Spark machines have the performance for large models. Windows 11 hardware therefore sorts into three tiers: PCs with no AI hardware, Copilot+ PCs, and RTX Spark-equipped machines aimed at developers and creators.

Demand is not guaranteed. Copilot+ PCs were not well received in the market, so RTX Spark machines may not catch on right away. Still, the analysis holds that as developer and creator machines they offer performance with no substitute, so some demand can be expected.

The deeper pressure is the cost of AI inference. Most AI functionality today is processed in the cloud, and as usage grows, equipment costs are becoming excessive. Even when users are pushed toward paid services, usage is hard to restrain once people understand what AI can do. The problem is common to businesses that provide AI, not just Microsoft. Shifting weight toward local inference on PCs, phones, and tablets is a natural response. Building SoCs from scratch takes time, so buying from SoC makers is the realistic route, and costs should fall as unit volumes rise. That would lower semiconductor costs for Microsoft too. In phones, extra memory and powerful NPUs may become standard.

The analysis expects layered market effects. In the short term, Windows ARM supply moves from Qualcomm alone to competition among several companies, widening OEMs' options. In the medium term, local inference capability may become a key spec driving PC replacement, with GPU and NPU hardware setting prices and differentiation. For chipmakers, AI investment that has centered on data centers would extend toward devices. The flip side is that some cloud inference revenue could migrate to the device. Microsoft's choice of an NVIDIA SoC, despite owning Windows, can be read as an effort to spread load off its cloud while rebuilding the device ecosystem. Whether other processor makers follow will be a touchstone for leadership in the AI PC era.