Product1 publisherNot yet confirmed elsewhere3 min readPublished
Microsoft sets Nvidia's first Arm laptop chip against the MacBook Pro at $2,599.99 and up
Microsoft's Surface Laptop Ultra, the first machine on Nvidia's RTX Spark Arm chip, starts at $2,599.99 and reaches $5,899.99 with 128GB of memory. Microsoft built it for developers running AI models locally, and so far its case against the MacBook Pro rests on the company's own performance claims.
The Product Desk
What happened
- Surface chief Brett Ostrum said Microsoft's 2024 Copilot Plus laptops, built first on Qualcomm chips, were aimed at the MacBook Air.
- Microsoft says the RTX Spark can run compressed 280-billion-parameter AI models on the laptop itself, models that needed the cloud a year ago.
- Putting CPU and GPU on one chip let Microsoft shrink the motherboard, fit larger fans and give more of the interior to battery.
- The Verge's reporter briefly played an emulated Gears of War: E-Day and found it smooth, with no slowdown when switching from wall power to battery.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- precedent Arm Windows now has a machine Microsoft explicitly built to take on the MacBook Pro, so from this launch on the platform will be judged on Pro-class work.
- constraint Teams whose daily software lacks Arm builds carry an emulation risk that a single short game session cannot size for them.
- exposure Early buyers take on first-generation risk in Nvidia's RTX Spark line, with performance backed so far by Microsoft's own claims.
The first real decision on the Surface Laptop Ultra is how much memory to buy. The 128GB top configuration costs $3,300 more than the 24GB entry model [15]. That is about 2.3 times the price for about 5.3 times the memory [16]. The RTX Spark is a single chip with both CPU and GPU, using unified memory, the approach Apple takes in its M-series [6]. For anyone buying this to run models locally, I'd treat the memory tier as the purchase. The launch coverage does not say which configuration ran the 280-billion-parameter models Microsoft cites [17], and it includes no measured comparison with a MacBook Pro.
Microsoft's list of buyers puts developers who want a strong GPU for local AI first. Creatives and professionals who would otherwise buy a MacBook Pro come second [7]. Jit Hirani, lead designer for Surface devices, said the team built the machine for itself, coming to work to build a product for its own use [8]. Product teams tell themselves this often, and it works when the team resembles the buyer. The creative buyer coming off a Mac is further from that team. For that person, The Verge reported the GPU "should" handle Blender objects with ease and cruise through Premiere and Photoshop work [18]. A creative lead will want to run one real project before trusting the hedge.
The MacBook Pro bar is Microsoft's own. Of the Qualcomm-era chips, Brett Ostrum, Microsoft's corporate vice president of Surface, said, "We knew that was not going to reach MacBook Pro-level compute." [3] Of Nvidia, he said, "It literally made a ton of sense for us in terms of how to go and compete with the MacBook Pro." [4] Because the Ultra is the debut platform for Nvidia's Arm PC chips [1], it is the first machine on which that bet can be measured.
Emulation is the quieter constraint. The RTX Spark is an Arm chip, so most games are not native and run emulated [11]. The Verge's reporter sensed Microsoft treading carefully about calling the Ultra a gaming laptop [14]. Games are the visible case. Any tool a team depends on that lacks an Arm build goes through the same emulation.
For the person who supports a fleet of these, the useful details are inside the case. No components are glued down, and disassembly instructions and QR codes are marked throughout [10]. The ports include HDMI and an SD card slot [13].
Sort the purchase on two lines: whether the team's software runs natively on Arm Windows, and whether the work needs large AI models in local memory.
Native software, large local models: this is the buyer Microsoft describes [7]. Spend on memory once Microsoft says which tier its model claim used.
Native software, no local models: the entry model is a polished Windows laptop whose GPU will spend most of its time idle.
Emulated software, large local models: pilot one unit on the team's real tools before ordering a second.
Emulated software, no local models: nothing in Microsoft's pitch is aimed at this buyer [7].
What to watch
- Independent benchmarks running the Surface Laptop Ultra and a MacBook Pro on the same local-model and creative workloads.
- Microsoft stating which memory configuration runs the compressed 280-billion-parameter models it cites.
- Whether Windows PC makers beyond Microsoft ship RTX Spark laptops, and at what prices.