AI hardware was the headline at Computex 2026. NVIDIA unveiled RTX Spark, a new chip platform built to run serious AI directly on a laptop or desktop, and Microsoft announced the Surface Laptop Ultra as one of the first machines to use it. The pitch is that the AI runs on the device in your hands, not in someone else's cloud. That is a genuine shift, so it is worth understanding what it actually buys a small business, and what it does not.
What was announced
Three things sit behind the same "Spark" name, and they are easy to confuse:
- RTX Spark is NVIDIA's new platform for Windows laptops and compact desktops. The chip pairs an Arm processor with a Blackwell graphics engine and 128GB of shared memory, and NVIDIA quotes around one petaflop of AI performance. The point of all that is to run capable AI models on the device itself. These systems arrive in the fall of 2026 from Dell, HP, Lenovo, ASUS, MSI, and Microsoft Surface, and they fall under Microsoft's Copilot+ PC label.
- The Surface Laptop Ultra is Microsoft's flagship example: a 15-inch laptop built on RTX Spark with 128GB of memory and a high-end display, rated at over 100 TOPS of AI performance (a measure of how much AI work the machine can do). It also ships in the fall.
- DGX Spark is the desktop cousin, a small "personal AI supercomputer" that has been shipping since late 2025 at around 3,999 US dollars. It can run and fine-tune large AI models with up to 200 billion parameters locally, which is developer and research territory more than everyday office work.
For context, these sit a notch above the ordinary AI PCs already on shelves. Today's mainstream Copilot+ laptops, the Intel and Qualcomm models, carry a neural processing unit rated around 45 TOPS, enough for the AI features built into Windows. The Spark machines aim much higher so they can run full models on their own.
What "AI-ready" actually means
Strip away the branding and there is one real idea: the AI runs on hardware you own, instead of on a server you rent. An "AI-ready" device has a chip designed to do that work efficiently, so tasks like transcription, summarizing documents, image editing, search, and assistant features can happen on the laptop itself.
Think of it as a spectrum rather than a yes or no. At the everyday end is a Copilot+ AI PC: a normal business laptop with a capable AI chip, fine for the AI built into Windows and Microsoft 365. At the high end are the Spark-class machines, which can host a real AI model in-house without sending anything to a cloud provider. Most businesses live at the everyday end and only a few need the high end.
Where the value is for a small business
On-device AI is not just a spec-sheet bragging point. For a small business, running AI locally has a few practical upsides:
- Sensitive data can stay on the machine. This is the big one. The most common AI risk for small firms right now is staff pasting client data into public chatbots, the problem we covered in shadow AI and data leakage. A model that runs locally is one way to give people the AI help they want while keeping confidential information on hardware you control.
- Predictable cost for some workloads. Cloud AI is usually billed per seat or per use. For steady, heavy use, owning the hardware can be cheaper and easier to budget than a growing monthly bill.
- It works offline and responds fast. No connection required, and no round trip to a distant server, which matters for quick, repetitive tasks.
- It future-proofs a refresh. If you are replacing aging laptops anyway, choosing AI-capable ones means you are ready for the AI features arriving in Windows and Microsoft 365 without buying twice.
Where the hype gets ahead of the need
Here is the honest part. A 1,000 dollar AI laptop for each person, let alone a 4,000 dollar Spark desktop, is not something most small businesses need to rush out and buy. The everyday Copilot+ AI PCs already on the market cover what the average team will actually use. The Spark-class machines earn their price only for specific jobs:
- A firm that handles confidential client data, for example legal, accounting, or healthcare, that wants to run an AI model entirely in-house rather than send anything to the cloud.
- Real AI development or data work, where you are building or fine-tuning models, not just using them.
- Heavy creative or engineering workloads (video, 3D, simulation) that benefit from the graphics power regardless of the AI angle.
If none of those describe you, the right move is not to chase the flashiest machine. It is to factor AI-readiness into your normal hardware refresh and otherwise keep your money.