Microsoft announced Surface Laptop Ultra on October 7, 2026, as a Windows laptop built around NVIDIA RTX Spark and marketed for local AI, graphics and other demanding workloads. Microsoft says local-agent workflows can use models with up to 120 billion parameters, while its business specifications list shared-memory configurations up to 128 GB. Those headline numbers depend on the configuration and workload.
Microsoft introduces Surface Laptop Ultra for local AI
The 15-inch laptop brings Microsoft’s Windows platform together with NVIDIA’s RTX Spark hardware. The event presentation shows the laptop on stage and covers its introduction. NeoTeo previously previewed the October 7 Windows and AI event.
Microsoft’s business listing gives two NVIDIA RTX Spark N1X options: an 18-core CPU with a 5,120-core Blackwell GPU, or a 20-core CPU with a 6,144-core Blackwell GPU. The configuration matters: the listing offers 24, 32, 48, 64 or 128 GB of LPDDR5x unified memory, shared between the CPU and GPU. The GPU cannot necessarily address the entire system-memory pool.
What the memory and FP4 claims mean
Unified memory gives the CPU and GPU access to a shared pool, rather than assigning each a completely separate memory amount. That makes the total capacity relevant to local AI workloads, but the system’s listed memory is not all necessarily available to the GPU.
Microsoft lists peak AI compute of up to one petaflop in FP4 for select configurations. FP4 is a low-precision number format used in AI computing. Microsoft describes the figure as theoretical and based on sparsity, a technique that can reduce computation by treating some values as zero. It is a peak specification, not a measure of how quickly a particular application or model will run.
Microsoft also markets local-agent workflows with models up to 120 billion parameters. That is a company-stated capability, not a promise that every configuration will run every model or that all such workloads will run at the same speed.
Local AI can still involve the cloud
Local AI means some model processing can happen on the computer, but Microsoft says not all models or workflows run locally and some AI workloads may require cloud services. NVIDIA describes OpenShell as a runtime that can route queries to local or cloud models under user-defined policies. The setup therefore can use both on-device and cloud processing, depending on the workflow and policy.
Software compatibility is another practical consideration. Surface Laptop Ultra runs Windows 11 on an Arm-based platform. Microsoft says many compatible Arm apps run natively; other applications may use Prism emulation. Specialized hardware drivers, plug-ins and older utilities may need compatibility checks. CUDA, NVIDIA’s accelerated-computing platform, supports GPU workloads, but software support still depends on the app and task.
U.S. prices and scheduled release
Microsoft’s U.S. consumer listing starts at $2,599.99. Its separate Surface Laptop Ultra for Business listing starts at $2,749.99 and schedules release for October 16, 2026, at 12:00 PM. The two prices refer to different listings, and the scheduled date applies to the business listing.
A historical account published October 8 attributed to NVIDIA CEO Jensen Huang a link between the expansion of Windows PCs and the early-1990s graphics opportunity pursued by NVIDIA’s founders.
