Why Your Next Computer Won't Need a GPU

Shops sell laptops with graphics cards as AI machines. The pitch is that AI needs a big card, a big power supply, and a loud fan. For most people it does not.

When you use ChatGPT, Claude, Gemini, or DeepSeek, the model does not run on your desk. It runs in a data center. Your computer sends the question and shows the answer. A browser and a stable connection are enough, and a cheap mini PC, a tablet, or a phone has both.

Small models now run on ordinary laptops too. At KOOMPI we have been testing an AI sidebar on the desktop that runs Google's Gemma on the laptop itself, with no graphics card, only the chip the laptop came with. The small version answers fast enough to read along. It needs no account and no connection, and the questions never leave the machine.

It has limits. The small model struggles with long tasks, and when we give it too many tools at once it starts dropping digits from numbers. So the sidebar keeps simple and private work on the laptop and sends heavy work to a cloud model.

That is the computer most people will own: a small model on the device for everyday private work, a large model in the cloud for the rest. Neither needs a graphics card.

This matters in Cambodia. Most businesses and schools here cannot spend thousands of dollars on AI hardware. They have phones, cheap laptops, and internet. A coffee shop in Kampot can use the same models as a company in Singapore. A school in Takeo can give students an AI tutor on the computers it already has.

The cloud half carries a risk. When the model lives on someone else's server, they set the price and the rules, and they can change both. Open models are the answer: if you can run a model yourself, the company serving it cannot hold you. That is why we work on the local half, even when the cloud is cheaper.

Before buying a new machine for AI, learn to use the one you have.