Microsoft Opens Pre-Orders for $2,599 Surface Laptop Ultra for Builders
Microsoft has opened pre-orders for the Surface Laptop Ultra, its first builder-class PC, starting at $2,599. The company uses the term "builder class PC" to designate the category. Engadget
Microsoft scheduled its Windows and Surface announcement for 10 a.m. PT on Wednesday, Oct. 7. Microsoft describes the Ultra on its official Surface page as "The performance laptop made for the world makers" and as built for world makers and creative pros who require cutting-edge AI performance, power, and portability.
The Ultra is built around a 15-inch Mini-LED PixelSense Ultra touchscreen with a 3:2 aspect ratio and up to 2,000 nits of peak brightness. Microsoft lists the panel at 262 ppi. Microsoft Mini-LED is a backlight design for higher contrast, 3:2 is a taller screen shape that preserves vertical code, timeline, and document space, and ppi means pixels per inch. Brightness at that level covers outdoor and high-ambient office use.
The haptic touchpad, which simulates clicks with vibration, is 30 percent larger than a typical 15-inch Surface notebook. The chassis includes USB-C, USB-A, HDMI, 3.5mm audio ports and a full-size SD card reader. It also introduces Magnetic Connect, combining magnetic charging with a USB-C jack.
Compute is based around NVIDIA's RTX Spark chip, with support for up to 128GB of unified memory, a single pool shared for processing and graphics, and up to one petaflop of FP4 AI compute, a speed measure using a compact format for AI math. Microsoft says the configuration can run local frontier-class models with up to 120B+ parameters, a count that reflects AI model size. Windows Central Large shared capacity paired with low-precision speed decides which models load locally and how fast they respond.
Microsoft says the Ultra has 2.5 times more thermal capacity, or ability to shed heat under sustained load, than any previous Microsoft notebook. It also says the machine delivers 99 percent of its max performance when not plugged into the wall.
Microsoft claims the Ultra delivers 6.2 times faster video generation than a MacBook Pro with M5 Pro chip, 4.3 times faster image creation than the same machine, and is 2.1 times faster to the first token. Time to first token, the delay before an assistant starts answering, is now a primary UX metric for assistants and agents.
Worth flagging, vendor cross-platform claims of that kind need independent verification across identical prompts, runtimes, and quantization, the process that compresses models to run faster. Full speed off the charger is also uncommon in high-power notebooks, so the 99 percent figure needs the same independent testing.
The broader context here is the shift from cloud-only inference, running AI on remote servers, to capable local inference for daily work. A 120B-parameter envelope on a portable machine changes iteration speed for prompt engineering, evaluation, distillation, and fine-tuning workflows that previously required a workstation or remote cluster. Local execution also keeps proprietary code, media, and datasets on device during early experimentation.
In my view, the most consequential specifications are memory capacity, FP4 throughput, and unplugged consistency together rather than in isolation. Peak TOPS figures, a rating of AI operations per second, have limited value if the memory pool cannot hold the model or if speeds fall on battery. Microsoft is tying those three variables together with the 128GB option, the petaflop FP4 rating, and the 99 percent off-wall figure.
Looking at what this means for teams shipping AI features, port selection and thermal headroom may prove as practical as token speed. Full-size SD ingest, HDMI output, and USB-A continuity reduce friction for field capture and lab hardware. Higher thermal capacity extends the window for long compiles, local evals, and batch image and video jobs without throttling. If delivered as specified, the result is a portable system that lets makers stay in flow without scheduling around power outlets or upload queues.
