From Developers to Everyday Users: What the New Surface Shows About AI on Your PC
How a chip that brings everything together helps a computer work on its own, and why Microsoft showed it today.
A Photo Gallery, Without Leaving Your Computer
Imagine you’re a photographer just starting out, and you’ve just finished a photo shoot. To deliver a professional gallery to your client, you would normally need to pay for a service to resize the images and pull together several tools to build the page.
In today’s presentation, Microsoft showed another way to do it. The user simply described what they wanted: resize the photos and create a local gallery. Perplexity’s AI agent found the images, checked the tools available on the PC and built a browsable gallery of 12 photos, with titles and details for each image. According to the presentation, the model ran on the computer itself without using any Perplexity credits.
But how can a computer do this on its own? The answer starts with the hardware inside the machines Microsoft made available for preorder today in partnership with NVIDIA: the Surface Laptop Ultra and the Surface RTX Spark Dev Box, a mini PC aimed at developers. Both use the same chip, RTX Spark.
One Chip in Place of Several Components
Computers have become part of everyday life, and most of us have heard of CPUs, RAM and GPUs. RTX Spark is a system on a chip (SoC): it brings the CPU and GPU together in a single package where they share the same memory.
To understand the difference, think of a traditional PC with a dedicated graphics card. The CPU, system memory and GPU are separate components connected through different interfaces and all communication between them runs through traces on the motherboard. When an AI agent performs many steps in sequence, moving data back and forth takes time and uses energy.
This idea isn’t entirely new. Some processors, known as APUs, have included an integrated GPU for quite some time. Apple develops its M series, while AMD has its Ryzen AI Max line. NVIDIA was already producing Grace Hopper and Grace Blackwell superchips for servers and data centers, and it also offers DGX Spark, a mini PC for developers. It is now bringing that approach to personal computers with RTX Spark.
There is also a practical effect on size: bringing components together can, in principle, make devices more compact and improve energy efficiency.
Two Buildings and One Warehouse
Imagine that a traditional computer has two buildings: the CPU and the GPU. The CPU uses system RAM as its main warehouse, while a dedicated GPU has its own warehouse, VRAM.
In some operations, when one building needs information stored in the other’s warehouse, that data has to be moved. It’s like sending trucks loaded with information between the buildings, creating traffic and using time and resources.
In chips with unified memory, such as RTX Spark, the CPU and GPU access the same warehouse. This reduces the need to copy data from one memory pool to another and lets both components work more directly with the same information.
Why This Matters for AI
RTX Spark offers up to 128 GB of unified memory in its highest-end configurations. In practice, instead of keeping one portion of memory exclusively for the CPU and another exclusively for the GPU, the system has a large pool that both use as needed.
This matters because the more memory the GPU can access, the larger the AI models the computer can run locally and the more tasks it can keep running at the same time. According to NVIDIA, that amount of memory allows models with up to 120 billion parameters to run locally. Put simply, the PC has room for a more capable AI running inside it, not just in the cloud.
Where Copilot Fits In
At the same presentation, Microsoft announced that it will bring hybrid intelligence to Copilot. Broadly speaking, hybrid means splitting the work: what can be handled on the device stays there, and the cloud steps in when the task needs more. Microsoft said the experience will use local AI, though how the work will be split in practice remains to be seen.
The Perplexity demonstration offers an example of what this could mean for users. Microsoft also mentioned agents such as Hermes, Perplexity and OpenClaw running on Windows, and introduced a new setup experience that promises to install these agents in just a few clicks, without a terminal or complicated configuration. Its stated ambition is to bring secure AI agent experiences to the billion people who use Windows every day.
A First Step
For now, the technology is still closer to developers than to everyday users: the Surface Laptop Ultra starts at $2,599, while the Dev Box starts at $5,999. But the message from the presentation is that the move toward everyday use has begun. Chips that bring the CPU, GPU and memory together, agents that perform tasks on the computer itself, and a Copilot that combines local AI with the cloud are the first pieces to bring AI agents into the daily lives of people who don’t code.
