AI Might Be More Affordable Than You Think

Employing AI in your business might be much more affordable than you think.

Almost every company using AI today is “renting” access. Every question you ask and every task you feed it travels to an outside provider’s servers, you pay each time, and the bill grows as the work grows.

There is good news on the horizon: open source models.

You run them on your own systems, behind your own walls, for a cost you control. And your data never leaves the building, so you have complete control over data privacy and security.

Until recently it wasn’t a real choice, because the local model option was too far behind. In other words local models just couldn’t solve hard problems.

That is what’s changing. The local models have become very good. The easiest way to think about it is they can be 6-12 months behind the frontier models. This means they have become “good enough” or soon will be for most tasks.

So the question changes. Not whether you can afford the best AI, but whether your work actually demands it. For the hardest problems and real R&D, “renting” still earns its price. For most of what a company does day to day, local models are quickly approaching “good enough”, and getting better every month.

A few notes:

• Running an open source model is not free. You still need a “big enough computer” to run it. At meaningful volume, the on-going cost difference is substantial.
• I am glossing over a lot of nuanced details required to make this work company need by company need. It does require time and effort to set this up correctly.