
For years, headlines about artificial intelligence have followed the same pattern: models are being trained faster, on more hardware, than ever before. The headlines are right. Epoch AI, a research group that tracks the computing power used to train notable AI models, estimates that training compute has been growing by roughly four to five times every year. That growth happens in purpose-built facilities filled with specialized accelerators, run by a handful of very large organizations. None of it happens on a small business's hosting plan. And yet the same trend has quietly changed what a small business can do with an ordinary server. As the frontier races ahead, capable models are also getting much smaller. Stanford's AI Index 2025 reports that in 2022, the smallest model to score above 60% on a widely used knowledge benchmark had 540 billion parameters; by 2024, a model with 3.8 billion parameters cleared the same bar. That is a 142-fold reduction in two years. Smaller models, cheaper AI services and mature open-source tools mean that AI hosting for small business is no longer a contradiction. A retailer can add semantic product search. A property manager can make ten years of maintenance notes searchable in plain language. An accountant can run a nightly forecasting job. Some of that work belongs on external AI services; some of it runs comfortably on a Linux virtual private server; a little of it genuinely needs more. This guide sorts those workloads out. It explains the difference between training and inference, the four kinds of AI work small businesses actually run, how to size CPU, memory, storage and GPU for each, how to run AI alongside a website without slowing it down, and where your data goes when you choose web hosting in Canada for AI features. It deals with AI that runs on your server. For the other side of the coin, the load that AI crawlers now put on your hosting, see our separate guide. What does "AI hosting" mean for a small business? AI hosting, for a small business, means running the parts of an AI feature that live on your own infrastructure: the application that calls...


