Why startups are building their own AI from scratch

A coding platform called Base44 has decided to stop relying solely on these third-party engines. They have begun building their own custom model—the mathematical software that processes information and makes predictions—with the goal of eventually making it perform better than the current industry-leading models, which are the most advanced, powerful engines available to the public today.

Think of these widely available models as a massive, pre-trained chef who already knows how to make every standard dish in existence. To make it "pre-trained," tech companies fed that AI millions of books, articles, and websites so it could learn the patterns of human logic. To make it better, Base44 is essentially hiring their own chef and putting them through a custom, rigorous culinary school. This requires massive amounts of computing power—which means thousands of specialized computer chips working in unison for weeks or months, constantly calculating millions of tiny math adjustments to teach the AI the nuance of coding. It takes enormous effort and electricity to show the software enough specific examples for it to develop its own specialty. If they succeed, they stop paying "rent" to big tech companies and create a unique product that no one else can replicate.

This shift highlights a growing tension in the tech industry: is it better to be a house guest or a homeowner? By owning their own model, Base44 gains full control over the quality and functionality of their product, which is a major advantage in a crowded market. However, they are also betting that they can compete with the deep pockets of the biggest tech firms. If they cannot train their model to be at least as smart as the industry leaders, they risk spending a fortune on a product that is ultimately less capable than the alternatives their customers could be using elsewhere.