A world full of Tokens

Anyone working with AI today produces a new type of consumable: tokens. Every word, every prompt, every response is broken down into these digital units, processed and billed. They are the invisible currency of AI and well on their way to becoming the new status symbol of knowledge work.

Thursday morning, LinkedIn, third scroll over my morning coffee. A guy with an open shirt collar posts his token dashboard. 847,000 tokens in a single workday. Below it: "If you're not burning at least 500k tokens a day, you haven't understood what’s happening." 214 likes, 38 comments, half of them fire emojis, the other one are rockets. I look at my own Claude window and find myself wondering whether this hustle dude is way ahead of me and of course why LinkedIn is serving me this content.

LookMaxxing, SleepMaxxing, MoneyMaxxing. If you're on TikTok or Insta, you know the logic: take one aspect of your life and optimize it to the max. Since spring 2026, there's a new contender, more on the LinkedIn side of things: TokenMaxxing. Professional context, but the ego level stays the same.

TokenMaxxing describes the trend of deliberately maximizing your AI token consumption to signal productivity. The more tokens flowing through your projects, the more valuable your work appears. In theory, at least.

But what are tokens, exactly?

A token is like the byte of language processing. Not a word, not a letter, but something in between. The word "artificial intelligence" gets broken down into several tokens by the model. "And" is a single one. So is a comma. In English, one token roughly equals three quarters of a word. This text you're reading right now has about 2,500 of them.

Every AI model has what's called a context window. Think of it as a desk: everything on it, the model can see and process. Everything that falls off ceases to exist. Your prompt, the system instructions, the conversation history and the model's response all need to fit on this desk at the same time. Input and output share one budget. Once the desk is full, the AI forgets the beginning of the conversation.

That's why longer prompts don't automatically produce better results. A clearly structured context of 10,000 tokens is often more useful than a chaotic one of 500,000.

Context windows have grown rapidly in recent years. In 2020, GPT-3 could process around 4,000 tokens. Today, mid-2026, five leading models each offer a million. That's more than the full text of War and Peace.

But more context doesn't automatically mean better answers. Research shows that language models process information at the beginning and end of the context best. What sits in the middle tends to get lost. You could say AI has the attention span of a meeting right after lunch.

The Digital West's Token Rush

What does any of this have to do with the workplace? In April 2026, the New York Times reported on an internal leaderboard at Meta called "Claudeonomics." An employee had set it up on the company intranet. It tracked the token consumption of over 85,000 employees. The top 250 received titles: "Token Legend" for the top spot, "Session Immortal" for particularly long sessions. The leader had consumed 281 billion tokens in 30 days.

Jensen Huang, CEO of Nvidia, has publicly said he would be "deeply concerned" if an engineer earning $500,000 a year wasn't consuming at least $250,000 worth of tokens. In some companies, token consumption is now being discussed as a productivity indicator and partly factored into internal performance reviews. Tencent gives its employees an annual token budget equivalent to roughly $30,000. Nvidia plans to provide engineers with tokens worth half their base salary.

If this continues, tokens will become the new status currency of Silicon Valley. Somewhere between stock options, minutes in the ice bath and the oat milk lattes per day.

This is reminiscent of the "lines of code" metric of the 1990s, when IBM evaluated developers by how many lines of code they produced. The logic was wrong then. It's wrong again today. Consuming a lot of tokens doesn't automatically make you productive. Still, a new way of measuring performance is emerging that we shouldn't underestimate. Whoever can generate stronger commercial output with a given token budget than their colleagues will have a strong hand at the next salary negotiation.

What this means for you and me

You see them everywhere on LinkedIn right now: AI-generated images placing a human employee next to an AI column. On the left, a person with an $85,000 annual salary. On the right, $85,000 in tokens. And the rhetorical question: Who would you hire?

This is where it gets uncomfortable for all of us. Because the math will keep shifting in favor of the machine. Projections suggest token consumption will increase 24-fold by 2030, to 120 quadrillion tokens per month. Chip costs are dropping 60 to 70% annually, but volume growth is outpacing cost reduction two to one.

Token prices are falling, fast. Google, OpenAI, Anthropic, they're all cutting prices on a quarterly basis. What costs $100,000 in tokens today could be $2,000 in two years. In four years, maybe five dollars. The salary on the left side? That stays. And tends to rise.

The uncomfortable question is: will our jobs be replaced by extremely cheap AI compute? For research, analysis, first drafts, communication, data preparation, the answer will increasingly be yes. The list grows by the day.

What will matter is agency. Someone has to define what's needed. Someone has to ask the question before the AI can give an answer. Someone has to decide whether the answer is good enough or leading astray. That's not a formality. It's where the actual value is created.

Token consumption may explode. But without the person who knows what the tokens are being used for, it's just expensive noise. That's where things stand in July 2026. The future doesn't belong to those who consume the most tokens. It belongs to those who best know what for.

tl;dr

Also July 2026: on one of the biggest Reddit subs, the question of what someone could say on a first date that would make you get up and leave immediately. The most popular answer by far: Maxxing.

And yet, the tokens and the maxxing remain. Because in the end, both promise the same thing: the biggest possible outcome.

With that in mind, happy summer with some Tokenminimizing, yours, Markus.