Surface Laptop Ultra price
At a San Francisco event, Satya Nadella and Jensen Huang opened preorders for a 15-inch AI powerhouse that runs 120-billion-parameter models with no cloud — and moved the compute bill from Microsoft's data centers to customers' desks.
Microsoft opened preorders on Wednesday for the Surface Laptop Ultra, and the Surface Laptop Ultra price — $2,599 to start — undercuts Apple's 16-inch MacBook Pro by roughly $400 at the entry level. Unveiled at a Windows and Surface event in San Francisco attended by Microsoft CEO Satya Nadella, Windows and Surface chief Pavan Davuluri, and NVIDIA CEO Jensen Huang, the 15-inch laptop is the first major Windows machine built on NVIDIA's RTX Spark platform, announced in June 2026. The machine begins shipping October 16, alongside a $5,999 Surface RTX Spark Dev Box desktop aimed at developers running open-weight AI models.
Why this matters: the data center moves to the desk
The strategic point is the one Reuters' Stephen Nellis put bluntly: work Microsoft currently performs inside its costly Azure data centers can now shift to high-powered Windows PCs that customers buy outright. The hardware bill moves from Microsoft's balance sheet to yours — a margin story disguised as a product launch. Nadella called the launch “the birth of the next generation of Windows agentic platforms.” Huang, who took the stage with Nadella, said, “We have reinvented the computer as we know it.” The deeper significance is an inflection in local AI: the most capable models are no longer gated behind cloud subscriptions. If the software delivers, this is the moment the AI PC stops being a marketing label and becomes a machine class.
RTX Spark explained: what's actually inside the chip
Surface Laptop Ultra specs: RTX Spark 128GB unified memory
The RTX Spark pairs an NVIDIA Grace Arm CPU with a Blackwell RTX GPU on a single die: up to 20 CPU cores, 6,144 CUDA cores, and up to 128GB of LPDDR5X unified memory delivered at roughly 300GB/s over NVLink-C2C. NVIDIA claims up to 1 petaflop of FP4 AI compute — read that figure carefully. It is a vendor number based on FP4 TOPS with sparsity assumed, not an independent benchmark, and no third-party testing exists yet. What the 300GB/s unified bandwidth actually changes is model loading: with 128GB of memory visible to both CPU and GPU, a 120-billion-parameter model can sit entirely in memory without shuttling weights across a PCIe bus — the bottleneck that has capped consumer laptops at far smaller models. Full CUDA support means developers' existing GPU code runs as-is. The machine around it: a 15-inch mini-LED PixelSense Ultra touchscreen with up to 2,000 nits peak HDR brightness, a haptic trackpad, and — notably in 2026 — a full port selection: HDMI, USB-C, USB-A, an SD card slot, and a headphone jack.
The price ladder and the MacBook Pro problem
Surface Laptop Ultra vs MacBook Pro: memory changes the comparison
The base Surface Laptop Ultra price of $2,599 sits about $400 below a 16-inch MacBook Pro starting around $2,999 (M5 Pro with 24GB, per Notebookcheck) — but the comparison that matters is memory. Apple's unified-memory ceiling and its pricing steps are exactly what Microsoft is attacking: pre-event leaks from Roland Quandt estimated configurations from $3,000 to $7,000, and Barron's points to memory costs as the driver — 128GB of LPDDR5X is not a commodity line item. Put the numbers in context: you are buying the cheapest path yet to 128GB of GPU-addressable unified memory in a laptop, a figure that would cost far more in workstation hardware. The open question is how many of the people who need that ceiling will accept an Arm-based Windows machine to get it.
How we got here: from Computex tease to launch day
Surface Laptop Ultra preorder opens ahead of the October 16 release date
RTX Spark was teased at Computex in May and formally announced in June 2026, but it is the product of years of Microsoft–NVIDIA engineering work. The history that matters is Windows on Arm's Qualcomm years — machines that promised efficiency and delivered app-compatibility headaches — and the benchmark Microsoft must beat: Apple silicon, which proved Arm laptops could be both fast and software-compatible. This is Microsoft's biggest Windows event since 2024, and it is the hardware half of the company's local-AI-agent push, alongside this autumn's Copilot overhaul. Read more about Microsoft's Copilot revamp turning Office into an AI super app.
Who wins, who feels the pressure
The beneficiaries are easy to name. AI developers get a portable CUDA machine with a 128GB memory ceiling — the open-weight ecosystem (models like DeepSeek, which can be downloaded and run on any capable device) gains a flagship consumer target. Microsoft gets to offload compute costs from Azure onto customer hardware. NVIDIA gets a beachhead in Windows PCs, one of the last major markets still dominated by Intel and AMD. The pressure falls on Intel and AMD, on cloud GPU rental vendors whose per-hour pricing now competes with a one-time $2,599 purchase, and on Apple, whose premium pricing room narrows when a credible local-AI alternative undercuts it by $400. Investor appetite for NVIDIA compute extends well beyond PCs — witness the scale of recent financing commitments for NVIDIA chips in SpaceX's $40 billion NVIDIA chips plan.
What the skeptics say
Uncertainty first: the 1-petaflop FP4 figure is vendor math with sparsity baked in, and independent benchmarks do not exist — treat every performance claim as marketing until third parties test it. Battery life was conspicuously absent from the announced numbers; for a machine pitched as the desk-side data center, efficiency claims need verification. Arm app-compatibility caveats are the ghost of the Qualcomm years — Microsoft and NVIDIA will be judged on emulation quality, not launch slides. And local AI is not total AI: open-weight models like DeepSeek can be downloaded and run anywhere, but proprietary models like ChatGPT and Claude remain app- or web-only, so the cloud is not disappearing.
The rest of the event: Windows gets agentic
Surface RTX Spark Dev Box price: $5,999
The laptop shared the stage with software. Microsoft announced Copilot developer updates including the ability to distinguish user actions from AI actions in every project and step — a small change with large implications for debugging agentic workflows. And in a notable platform move, Muse — Meta Platforms' personal AI agent — is coming to Windows, putting a rival's assistant on Microsoft's home turf. For developers who want the chip without the laptop, the $5,999 Surface RTX Spark Dev Box is the workstation play: a desktop built for running open-weight models locally. OEM partners are already rolling out their own RTX Spark machines, including the Lenovo and Acer RTX Spark AI PCs.
What happens next: breakout win or developer niche?
Two scenarios. The breakout: OEM laptops follow — Dell, ASUS and HP are the names to watch alongside the already-announced Lenovo and Acer machines — software vendors optimize for local inference, and the Surface Laptop Ultra becomes the reference design for a genuine MacBook Pro alternative. The niche: benchmarks disappoint, Arm compatibility stumbles, and the machine settles into a developer tool — valuable, but not a mass-market shift. Hanging over both is the security question Reuters flagged: containing long-running AI agents safely on personal computers. NVIDIA is working to prevent a repeat of the Hugging Face hub hack, and Apple is tightening Mac drive access — the agentic platform era will be judged as much on containment as on capability.