Surface Laptop Ultra Cracks Intel and AMD’s PC Chip Grip

Microsoft’s Surface Laptop Ultra arrived this week as the most powerful machine the brand has shipped, built around an NVIDIA Blackwell RTX graphics processor, up to 128GB of unified memory and one petaflop of on-device artificial-intelligence compute. The headline trick: it runs AI models with up to 120 billion parameters entirely on the laptop, with no cloud connection required.

The device is the consumer storefront for a much larger move. The same silicon inside it, NVIDIA’s new RTX Spark superchip, walked straight into territory Intel and AMD have controlled for three decades, and Wall Street noticed within hours of Jensen Huang, NVIDIA’s chief executive, taking the Computex stage in Taipei.

What Microsoft Put Inside the Surface Laptop Ultra

The laptop pairs a 15-inch mini-LED PixelSense Ultra touchscreen with internals normally reserved for workstations. Microsoft says the panel reaches up to 2,000 nits of peak high-dynamic-range (HDR) brightness, the brightest display it has ever shipped, at 262 pixels per inch. It sits in a precision-machined chassis offered in Platinum and Nightfall, with the largest haptic touchpad on any Surface to date.

The reason the machine exists is local compute. Microsoft quotes a full petaflop of AI throughput, a figure drawn from NVIDIA’s FP4 (4-bit floating point) math with sparsity switched on, and says the unified memory pool can be shared dynamically between processor and graphics chip depending on the job. Memory shifts to wherever a workload needs it, whether that is three-dimensional rendering, video editing, or several AI models running at the same time.

• 2,000 nits peak HDR brightness on a 15-inch mini-LED display, 262 pixels per inch
• 128GB unified memory shared dynamically between CPU and GPU
• 1 petaflop of AI compute using NVIDIA FP4 precision with sparsity
• 120 billion parameters, the largest model size the laptop can run locally

The hardware is tuned for Windows and the local AI features Microsoft has been threading through the operating system, the same on-device direction behind its on-and-off Copilot integration in Windows 11. Brett Ostrum, corporate vice president for Surface, pitched the device as a creator and developer tool rather than a mainstream notebook. Pricing, storage tiers and exact configurations were left blank in Microsoft’s Surface Laptop Ultra announcement.

The Shared Silicon Behind the Surface and Five Rivals

Strip away the Surface badge and the engine is NVIDIA’s RTX Spark, the company’s first PC processor in more than a decade. It fuses a 20-core Arm-based central processor with a Blackwell graphics chip carrying 6,144 CUDA cores, linked by NVLink-C2C (a chip-to-chip interconnect that moves data between the two parts without the usual bottleneck). NVIDIA rates the package at up to 300 gigabytes per second of memory bandwidth.

The chip was co-developed with Microsoft and MediaTek, and it is the consumer cousin of the workstation silicon NVIDIA showed at its developer event last year. It also runs the full CUDA (Compute Unified Device Architecture, NVIDIA’s parallel-computing software stack) natively, the first Windows laptop chip to run the full CUDA stack natively.

Microsoft is not the only customer. NVIDIA lined up a wall of laptop makers for a fall launch and expects the platform to span dozens of machines:

• Dell, HP, ASUS, Lenovo and MSI, each shipping their own designs alongside Microsoft
• Acer and GIGABYTE following the first wave
• More than 30 laptops and 10 desktops planned in total

That breadth is the point. A single Surface flagship is a statement; six brands and dozens of devices arriving together is a platform, and you can see the full roster of partner laptops and desktops at Computex already lining up behind it.

Intel, AMD and Qualcomm Lost Billions Before Lunch

Markets read the move at once. As Huang detailed the platform on June 1, shares of the incumbents it threatens fell hard while NVIDIA’s own stock climbed.

NVIDIA rose to roughly $224.27 a share, lifting its value to about $5.47 trillion, a scale consistent with the data-center growth in NVIDIA’s latest quarterly filing. The selloff in the other three was a bet that a serious new entrant had just appeared in a market they had effectively split among themselves.

Why CUDA on Windows Is the Wedge

Raw specifications are not what worries Intel and AMD most. The wedge is software. For two decades CUDA has been the default language of GPU computing, the toolset that AI researchers, video professionals and game developers already know, and until now it lived mostly on desktop graphics cards and data-center servers, not inside a thin Windows laptop.

Software That Already Has an Audience

Putting native CUDA in a portable machine hands developers a familiar workflow without a rebuild. A creator who trains a model on an NVIDIA server can run a trimmed version of it on the train home, on the same code path. That continuity is exactly the advantage Qualcomm has struggled to build for its Arm Windows chips, which lean on a newer and thinner software base.

Microsoft made the partnership the centerpiece of its sales pitch.

Engineered with NVIDIA from the silicon up … built on Windows.

That line, from Microsoft’s announcement, captures the strategy: hardware and software designed together rather than a graphics chip bolted onto a generic notebook.

An Arm Push Microsoft Has Tried Before

Windows on Arm has a checkered history, with two earlier attempts that never dented Intel’s grip. What is different now is momentum. Research firm Canalys expects Arm-based PCs to reach 30% of the market by 2028, up from about 10% in 2025, and a name as heavy as NVIDIA backing the architecture changes the math. It also fits the same local-AI thinking behind Microsoft baking Copilot AI directly into Edge rather than leaning only on the cloud.

The MacBook Pro in the Crosshairs

The comparison Microsoft wants buyers to make is with Apple. For years the MacBook Pro owned the high-memory, quiet, long-battery niche for creative professionals, largely thanks to Apple’s unified-memory design that lets the processor and graphics share one pool of RAM.

The new Surface copies that core idea and adds NVIDIA’s software base on top.

The difference Microsoft is banking on is reach. Apple’s AI tooling is strong but closed to macOS, while CUDA travels across a far wider base of Windows software, games and developer projects. Whether that breadth beats Apple’s years of battery and efficiency tuning is the open question the spec sheet cannot answer.

The Unknowns That Decide Whether This Sticks

The capability is settled. The economics are not. Microsoft published a long list of numbers and a short list of answers to the questions buyers actually ask.

Jason Tsai, an analyst at DigiTimes, cautioned that NVIDIA’s success may hinge less on the technology than on cost, and that complete systems probably need to land near $1,500 to reach beyond a niche audience. None of the partners has confirmed a price.

• Price: undisclosed by every partner so far; analysts peg viable systems around $1,500
• Battery: the all-day claim rests on internal testing of pre-release units and may vary by use
• Specs and regions: CPU details, configurations and regional availability are unannounced, and the product is pre-release, subject to regulatory approval
• App compatibility: Windows on Arm still relies on translation to run some older x86 software

Qualcomm, whose stock took the hardest knock, responded publicly by welcoming NVIDIA to the PC market, a sign the Arm-on-Windows fight is now genuinely crowded. Both Qualcomm and AMD have next-generation Arm parts of their own on the way.

If the laptops ship this fall near that $1,500 mark with battery life that survives a working day, NVIDIA turns one keynote into a real second front against Intel and AMD. If they arrive expensive, hot, or stumbling over Windows app compatibility, the most powerful Surface yet becomes a halo product the incumbents can simply wait out.

Frequently Asked Questions

When will the Surface Laptop Ultra be available?

Microsoft says the laptop will arrive later in 2026, in the fall window NVIDIA set for its hardware partners. The company has not given an exact on-sale date, and notes the unit is pre-release and dependent on regulatory approval in different regions.

What NVIDIA chip is inside it?

It runs NVIDIA’s RTX Spark superchip, which combines a 20-core Arm-based processor with a Blackwell graphics chip carrying 6,144 CUDA cores and a shared pool of unified memory, linked by an NVLink-C2C interconnect.

Can it run AI models without an internet connection?

Yes. Microsoft says the machine can run models with up to 120 billion parameters locally, using its on-device compute, so workloads like generation and rendering do not have to call out to a cloud server.

Will it run normal Windows software?

It runs Windows, but because the processor uses the Arm architecture rather than x86, some older applications rely on a translation layer. Native CUDA support is built in, which matters most for AI, creative and developer tools.

How does it compare with a MacBook Pro?

Both use a unified-memory design and top out around the same memory ceiling, but the Surface adds native CUDA on Windows, while Apple keeps its AI tooling on macOS. Apple still holds an edge in proven battery efficiency.