Updated October 2026
Surface Laptop Ultra vs MacBook Pro M5 Max for local AI
Two laptops now hold 128GB of unified memory. What's confirmed about the RTX Spark Surface, the one number NVIDIA hasn't published, and which to buy if you can't wait for reviews.
As an Amazon Associate I earn from qualifying purchases. Product links on this page go to Amazon or the Microsoft Store; both are affiliate links, and the price you pay is the same.
The picks
Microsoft Surface Laptop Ultra (RTX Spark, up to 128GB unified memory)
Order the 64GB or 128GB configuration; the 24GB and 32GB ones are the base Mac mini story in a bigger chassis. Pre-review pick: bandwidth and thermals are unverified until units ship on October 16.
CUDA in a laptop that holds a 70B model. Bandwidth is unpublished (about 300GB/s derived) and the Windows-on-Arm tools are new, so it's a pick for people who want CUDA on Arm and can wait for the software to settle.
MacBook Pro 16-inch (M5 Max, 40-core GPU, 128GB unified memory)
The known quantity: 128GB at 614GB/s, shipping now. Pick the 40-core GPU. If you don't need CUDA, this is still the laptop to buy.
The fastest laptop for running a 70B model in memory. The 128GB Strix Halo and RTX Spark laptops fit the same models at well under half the bandwidth. It will be loud and warm doing it, and it will do it on a plane.
Microsoft Surface RTX Spark Dev Box (128GB)
The same platform without the screen. If the laptop part doesn't matter, this or the DGX Spark is the better buy. Ships late November.
The same silicon as the laptop at 140W instead of 45 to 80W, so it should hold its speed under sustained load. If you don't need the screen, this or the DGX Spark is the better buy.
What changed
Until October 2026 one laptop held a 70B model in memory: the MacBook Pro with M5 Max and 128GB. On October 7 Microsoft opened pre-orders for the Surface Laptop Ultra, built on NVIDIA’s RTX Spark: up to a 20-core NVIDIA Grace CPU and a 6,144-core Blackwell GPU sharing up to 128GB of unified memory, running Windows with the full CUDA stack. First units ship October 16. Acer, ASUS, Dell, HP, Lenovo, MSI and Gigabyte have announced RTX Spark machines as well; this guide covers the Surface because it’s the one with a published configuration list and a ship date.
This guide is written from the spec sheets, before anyone has benchmarked one. The “Updated” date at the top moves when reviews change the picks.
What’s confirmed
- Memory. Up to 128GB unified, shared between CPU and GPU. The 20-core chip comes in 32GB, 48GB, 64GB and 128GB; an 18-core version reportedly starts at 24GB. Microsoft says the GPU can address less than the total depending on configuration, as on a Mac.
- Software. Full CUDA on Windows. That is the whole reason to consider this over the Mac: Ollama, LM Studio, llama.cpp, PyTorch and the fine-tuning tools all have CUDA paths first and Metal or ROCm paths second.
- Models. Microsoft’s own claim is “models up to 120B parameters” locally. A 70B model at 4-bit is about 40GB, so 128GB has room; 64GB runs 70B models the way a 64GB Mac mini does, which is badly.
- The rest. 15-inch 120Hz touch display, up to 2TB of user-replaceable storage, three USB-C, USB-A, HDMI, SD slot, under 4.5 pounds and under 18mm. The base configuration lands in the $1,000 to $3,000 tier; the 64GB and 128GB configurations that matter here are No ceiling.
The number nobody has published
Memory bandwidth decides tokens per second, and neither Microsoft nor NVIDIA has printed one for RTX Spark. The 600GB/s figure in NVIDIA’s announcement is the link between the CPU and GPU, not the speed of the memory itself.
What we do know: NVIDIA’s DGX Spark, which uses the same Grace-plus-Blackwell design with 128GB of LPDDR5X, is 273GB/s. If the laptop’s memory system matches it, the Surface sits with the Strix Halo boxes: it holds the same models as the MacBook Pro M5 Max and runs them at well under half the speed (614GB/s on the Mac). That is an inference from a sibling product, not a spec. Treat it as the likely case until a review measures it.
Mac or Surface
MacBook Pro M5 Max if: you want the speed, you want it now, and your tools run on Metal (Ollama, LM Studio and MLX all do, with nothing to configure). It’s shipping, it’s measured, and 614GB/s is the fastest memory in any laptop.
Surface Laptop Ultra if: your work needs CUDA. Fine-tuning, anything built on PyTorch with CUDA kernels, Windows-only pipelines, or a team that standardizes on NVIDIA tooling. You give up a known bandwidth number for a known software stack. Order 64GB at minimum; 128GB if 70B models are the point.
Wait if: you can. Review units will be measured within weeks of October 16, and the bandwidth question settles the whole comparison. The Mac isn’t going anywhere, and the Surface will still be there in November.
The desktop version
The Surface RTX Spark Dev Box is the same platform in a small desktop: 128GB, 2TB, Ethernet, and Windows with VS Code, Git, WSL, Python and GitHub Copilot preinstalled. It ships late November 2026. It’s the same class as the NVIDIA DGX Spark in the build guide, and it costs more. What the extra buys is Windows and a ready-to-code setup; the DGX Spark runs NVIDIA’s own Linux-based DGX OS. If you don’t need the laptop and are fine with Linux, the DGX Spark is the cheaper way to the same hardware class.