Token Dealer

Weekend Food for Thought on … Token Dealer. In my youth, I had bad company. Some of my friends were drug dealers. So, I’m familiar with the business model: the first time is free, the second time is a “friends” price, the third time you pay the regular price, and eventually, once you’re addicted, you pay the high price.

This is exactly the business model I see with OpenAI & Co. Your LLM providers are token dealers. Instead of paying per gram, you pay per token.

Getting started is free (aka “freemium”). Then you get heavily subsidized Pro/Plus/Premium/What-Ever-Poser prices. When your consumption spikes, you’ll pay the regular price—that is, per token. And once you’ve integrated GPT & Co. into your core business processes (aka “Agentic AI”) and made your company dependent on them, they’ll raise their prices.

In other respects, too, I consider many LLM providers to be token dealers:

Most of their chatbots are very pleasing: “Martin, that’s a good question.” That might not be addictive, but it makes you feel good. Just like drugs.

You need more and more. An employee has ChatGPT generate a presentation for a colleague, and that colleague prompts ChatGPT to generate a summary from those 100 generated slides. The workflow hasn’t become any more efficient or effective. But a lot of tokens were burned.

Some even practice tokenmaxxing: they brag about how many tokens they’ve burned. It’s similar to what you see at Oktoberfest with alcohol—people brag about how many liters of beer they’ve drunk.

Withdrawal symptoms are already starting to show. If Antrophic is offline for just one day, the developer can’t get any code done without Claude and posts about it on LinkedIn.

The dose has to keep getting stronger. The next high has to hit even harder. With ChatGPT or Claude, it’s the new versions that have to kick: faster, smarter, better.

The winner takes it all. Just like the big cartels: every LLM provider is trying to stake out its territory and hopes to drive the others out of business. In the end, there can only be one.

By the way, I also used an LLM for this post—so I consumed a few tokens, but in controlled doses. As with any drug, the dose makes the poison. In the right dose, it might even be medicine. Too much, and it’s poison.