The dawn of ‘Tokenmaxxing’ and the 2026 AI hangover
For the last three years, the corporate world’s relationship with artificial intelligence (AI) has been a lot like a toddler taking its first steps, or maybe a baby monkey trying to work out what to do with a camera.
Initially, there was The Era of Scaling and Praying (2022–2025), a typical phase after the invention of nearly every General Purpose Technology (GPT) since the railroad. During this period, tech giants invested hundreds of billions in infrastructure, creating huge engines even before knowing if their products would be useful or how they’d be operated.
Large Language Models (LLMs) followed, but the major breakthrough was The Age of Agents (Late 2025). Tools like Claude Code and OpenAI’s Codex transformed AI from simple chatbots into autonomous software agents. Rather than passively responding to prompts, these agents can loop, plan, execute, and self-correct. This shift changed the story from “Is anyone using this?” to “Anthropic’s servers are literally struggling under corporate demand.”
And now – mid -2026 – we have the hangover. Why a hangover? Well, simply speaking, the burning question isn’t whether AI can do the job – it’s how many can actually afford the bill.
Why agents are costing fortunes
A Large Language Model (LLM) or even a Small Language Model (SLM), takes a question and spits out an answer. An agent, however, runs an ongoing cognitive loop. It writes code, tests it, hits an error, updates its context, and tries again.
And every single step in that loop burns tokens (data processing units).
Asking an LLM a question consumes about 100 tokens for every 75 words it produces. However, instructing an AI agent to perform tasks like following a competitor’s announcements, tracking property opportunities, or alerting a fund manager to a new investment theme uses nearly 100,000 tokens before an answer is even given, according to U.S.-based SemiAnalysis. To put that in perspective, this is more than twice the length of the entire novel The Great Gatsby and roughly the same word count as Where the Crawdads Sing by Delia Owens or The Hobbit by J.R.R. Tolkien.
Tokens are essentially money, and companies are only now realizing this as their corporate AI token usage has soared. A typical business implementing AI consumes between 1 billion and 10 billion tokens each month. Year-over-year token consumption has increased approximately 10 to 13 times. According to MindStudio, the growth of autonomous AI agents means that a single approved user task can trigger a series of internal queries, amplifying token use by 8 to 15 times at the task level. Additionally, the WSJ reports that cloud giants like Google process over 3.2 quadrillion tokens monthly, marking about a sevenfold increase from the previous year.
In many cases, running the AI agent has become more expensive than hiring the team of human engineers it replaced.
Rise of the ‘Tokenmaxxers’
How did corporate spending get so out of hand, so early? The answer is Tokenmaxxing.
In the rush to prove they were AI-forward, some Silicon Valley giants – think Google and Meta – reportedly created internal leaderboards ranking employees by how many tokens they consumed. The result was a culture of reckless optimisation – workers threw massive, unchecked projects at agentic AI just to watch their numbers go up.
But massive token consumption hasn’t automatically translated to flawless productivity. Instead, it’s led to a digital phenomenon known as code churn.
Faros AI reported a staggering 800%+ increase in code churn (lines of code deleted vs. added) in environments with high AI adoption.
Instead of saving time, humans are spending their days cleaning up endless loops of broken, AI-generated code, leading to massive engineering burnout.
Is the bubble popping, or just calibrating?
If the boom in AI stocks has been based on the double digit revenue growth reported by AI operators, but that revenue growth is due to Tokenmaxxing, it is reasonable to conclude future growth rates will be lower.
Indeed, the spending backlash has been swift. Fortune 500 executives have quietly admitted to reporters that their companies have been burning through token budgets like a drunk gambler. Rumours are swirling that even early adopters like Uber and Microsoft have scaled back some of their autonomous agent licenses due to unmanageable costs.
Cognitive scientist Gary Marcus warned this week that if productivity disappointments continue to mount across the board, the entire AI bubble will pop.
A more realistic assessment might be that AI adoption will continue, but a correction in its pace is predictable.
Box CEO Aaron Levie famously pointed out that many CEOs fell victim to a temporary “AI psychosis” – the irrational belief that more AI is always better, and that every prototype an agent spits out is an instant billion-dollar idea.
As tech analyst Doug O’Laughlin recently noted, every disruptive technology undergoes a period of trial and error that includes some painful episodes. His theory is that after a new technology is released, it is met with under-spending and scepticism. This is followed by wild over-spending and reckless experimentation. Next is the indigestion phase – the dramatic, panicked pullback. Eventually, adoption rates settle at a stable equilibrium.
The 64-million-dollar question, of course, will be whether that equilibrium rate is fast enough to justify the valuations being suggested, for example, for the Initial Public Offerings (IPOs) of Anthropic and OpenAI.
But right now we are smack-bang in the middle of indigestion. The trillion-dollar bet on AI isn’t necessarily doomed, but the era of the blank cheque appears to be over.
Roger Montgomery is the Founder and Chairman of Montgomery Investment Management. Roger has over three decades of experience in funds management and related activities, including equities analysis, equity and derivatives strategy, trading and stockbroking. Prior to establishing Montgomery, Roger held positions at Ord Minnett Jardine Fleming, BT (Australia) Limited and Merrill Lynch.
He is also author of best-selling investment guide-book for the stock market, Value.able – how to value the best stocks and buy them for less than they are worth.
Roger appears regularly on television and radio, and in the press, including ABC radio and TV, The Australian and Ausbiz. View upcoming media appearances.
This post was contributed by a representative of Montgomery Investment Management Pty Limited (AFSL No. 354564). The principal purpose of this post is to provide factual information and not provide financial product advice. Additionally, the information provided is not intended to provide any recommendation or opinion about any financial product. Any commentary and statements of opinion however may contain general advice only that is prepared without taking into account your personal objectives, financial circumstances or needs. Because of this, before acting on any of the information provided, you should always consider its appropriateness in light of your personal objectives, financial circumstances and needs and should consider seeking independent advice from a financial advisor if necessary before making any decisions. This post specifically excludes personal advice.
