'It's a beast, but not at everything' — RTX Spark-powered Microsoft Surface Laptop Ultra lives up to expectations, but you need to be a big fan of local AI

Nvidia's RTX Spark laptops hit the shelves last week, led by Microsoft's Surface Laptop Ultra , and the chip supports up to 128GB of memory.

This laptop starts at $2,599.99, but does its performance justify the price tag? I took a look at some of the earliest in-depth reviews of the Microsoft Surface Laptop Ultra to ascertain what this device offers and whether it is worth every penny.

What reviews say The prowess of the RTX Spark laptop was acknowledged, but all the reviews agreed on one thing: it’s not foolproof and has some downsides.

“It's a beast, but not at everything,” said Zac Bowden of Windows Central , while Tom's Guide went further, calling it “not a MacBook Pro killer but something much weirder and wildly powerful.” The praise is concentrated in one area: local AI models.

Every outlet that tested them put the Surface ahead of Apple's M5 Max and AMD's Ryzen AI Max+ 395, the most powerful mobile chips for laptops.

Its gaming performance, however, did not exactly get much praise – while some said it is just okay, others think it’s a weak spot.

Price Most people would be better served getting an Intel or AMD laptop with a dedicated NVIDIA GPU instead, as these often cost less than the cheapest RTX Spark laptops right now Cost is the sticking point for RTX Spark.

Microsoft's Surface Laptop Ultra starts at $2,599.99 for the 18-core model with 24GB of memory and 512GB of storage.

The 20-core, 128GB, 1TB configuration tops out at $5,899.99, and the lead times on this model already stretch to late December.

The jump is steepest where the platform is strongest.

The 64GB to 96GB of GPU-accessible memory that lets 35B-parameter models run well only comes with the 128GB build, and a 32GB or 48GB Surface won't fit the same models.

Tom’s Hardware warns that “sky-high DRAM and NAND price hikes” will be a headwind for the platform in the near term.

That is a lot to spend, and you need a local AI workload to justify it.

The biggest challenge is the “eye-watering price tags,” since that money still buys “a lot” of cloud tokens from OpenAI and Anthropic, as Tom's Hardware puts it.