The concept of AI tokens can be likened to building bricks or shovels, where companies sell the tools needed to do fundamental work. This idea originates from the California gold rush, where not everyone found gold, but everyone looking for it bought picks and shovels. Similarly, AI companies’ Large Language Models (LLMs) can be thought of as shovel factories, producing smart shovels that can perform various tasks across writing, coding, research, and more.

As foundation model companies control these shovel factories and have access to effectively unlimited shovels, the question arises: how can startups, which have to pay for tokens and face rate limits, compete against them? The answer may lie in creating products that are incredibly niche or too risky for a general LLM company to touch. However, with the rapid integration of LLMs into various industries, including medical, legal, and finance, it seems like everything is on a timeline before it gets absorbed into LLM platforms.

The medical industry is already witnessing this phenomenon, where hospitals can use a single LLM to assist doctors, navigate health plans, handle scheduling, and more. While speed, focus, and trust might have been advantages for startups, these moats disappear when LLMs can throw unlimited resources at a problem. With the ability to run a startup that once required hundreds of people now possible with a small team, LLM companies can become multi-headed hydras, dominating every industry.

So, can patents and proprietary data protect startups from platform risk? Or is everything already on a clock, waiting to be integrated into LLM platforms? The future of startups in a world of unlimited AI resources remains uncertain, and it’s crucial to explore strategies for survival and growth in this new landscape.

Photo by Sahaban Nayak on Pexels
Photos provided by Pexels