AI Token

AI has proven its feasibility in many real-world tasks. In many cases, it exceeds the ability of an entry-level executive. AI runs on trained models. 

A trained AI model contains the "intelligence". To access the intelligence, we "speak" with the AI. The language we use to communicate with AI is token. Every conversation has been done in "token". 

AI model companies like OpenAI, Google, and Anthropic spent a lot of money training AI models. Each iteration of their models gets smarter. The key to commercializing their investment in training AI models is "token".They charge us based on tokens. Imagine AI model companies as the telcos; they spent money training the model (setting up the antenna and network stations). Then the AI model company charges us based on the token (the MB data we consumed, which the telcos charge us for).

The given token is the charging unit. In the long run, I believe AI companies that control the infrastructure will win because they are the cost center for training models, hence they have "ultimately" better margins on the cost of training and serving new AI models. Even if they are behind in the "intelligence" of the trained AI models, they can ultimately commercialize and reinvest the proceeds to improve their models.

Microsoft has been quiet on AI models. I believe Microsoft is betting on a strategy that makes AI models a commodity. AI companies like OpenAI sought to make their AI models proprietary and charge a premium for access. On the other hand, Microsoft might be betting on making AI models a commodity so that the end users switch between models.

If AI token is an infrastructure play and a cost game (imagine the data packet business for the telcos), then a company with an infrastructure culture, mindset, and cost structure will win. I am betting on AWS, Google, and Microsoft. On top of that, other players like Meta, Alibaba, X, and ByteDance will always be the deep, vertically integrated AI models in their respective spaces, and they will be strong contenders to generic AI models.

Anthropic and OpenAI might seek to kill SaaS companies by expanding into other SaaS categories. Existing SaaS companies are working on their AI features. Startups are creating an AI and agent-first SaaS; this is a new breed. Together, they represent the three main camps in the solution tier. It's interesting to see who wins in this territory fight. Which camp are you betting? Which camp are you in? Share your thoughts.