This article was originally published on Computing.co.uk and is authored by Simon Crichton, CEO, and David Savage, Technology Evangelist.

A new phrase has started to spread in tech circles: Token-o-palypse. This refers to the ‘apocalyptic’ costs of running AI models, which have spiked significantly in the last few weeks. Quite simply, is AI becoming too expensive?

At the core of token-o-palypse is the hidden cost of internal reasoning in newer AI models.

Historically, a prompt might use 100 input tokens and yield 400 output tokens, resulting in a predictable bill for 500 tokens.

However, newer models (like Opus 4.6) are designed with internal loops to stop and think before answering. They construct a reasoning plan, test their answers, iterate, and even create baby agents to find the best output.

While this delivers highly refined answers, the model is generating and consuming its own tokens in the background. This autonomous looping can burn through three to ten times the number of tokens per prompt.

Because these newer models are already more expensive on a unit-token basis, an interaction that used to cost 500 tokens can suddenly consume 4,000 to 5,000 tokens without the user explicitly asking for that level of reasoning, causing corporate token bills to skyrocket overnight.

Given that many businesses are already struggling to create return on investment from AI – the 2025 Harvey Nash Digital Leadership Report found that two-thirds of tech leaders are not realising ROI – this begs the question of whether the AI hype could begin to flounder due to hard financial realities. Will CEOs continue to sign off on the escalating AI budgets being asked for?

Managing costs through ‘Tokenomics’

The cost of tokens is unlikely to fall, but there are ways businesses can manage the bill.

Organisations are realising they need to be more strategic about the tools they deploy, which starts with understanding the problem they are trying to solve and how best to do it – AI may not always be the answer.

Then there’s the question of tools. The great majority of enterprise AI workloads do not actually require the latest, most expensive language models. Instead, businesses can control their token expenses and maintain functionality by shifting to smaller, open-source models, or by training their own internal models for specific tasks. As one tech leader we spoke to recently put it, “Everyone wants a Ferrari, but sometimes a bicycle will do the job.”

The easiest way to explain this in practice is to look at where AI is being deployed. In some instances, AI is being used instead of good old-fashioned automation.

Another approach is the use of ‘recipes’, which involves creating a playbook of AI workflows and calculations that can be taken and applied in different contexts – obviating the need to have AI recalculate what has already been done.

An additional focus is on combating agent sprawl or shadow AI, by ensuring there is a coordinated, centralised approach to AI deployment rather than a wild west where multiple departments start spinning up agents independently and adding to the token bill. Strong AI governance is becoming even more essential.

But while good tokenomics of this kind helps, they only goes so far. There is another fundamental question the token-o-palypse raises, which in our view urgently needs to be debated: does the balance between human and machine need to be reset?

A shift back towards people?

In the talent market, volume recruitment of traditional core IT roles such as developers and testers, especially at junior levels, has been impacted as companies look to do more with AI. However, in our conversations with tech leaders, there is evidence that some organisations are beginning to hire more junior talent again because of the elevated costs of AI.

At the repeatable task end, in many instances it is becoming more economical to use more junior and early careers staff than AI – with the same quality of end output.

Is token-o-palypse therefore the wake-up call that industry needs? The beginning, perhaps, of a restoration of a healthier balance?

Let us not forget all the capabilities and attributes that human professionals have, which AI struggles to deliver. Critical thinking based on experience and judgement, appropriate scepticism and the ability to stand back, strategic insight, human empathy and understanding.

Balancing the pipeline

There is also the key question of maintaining a talent pipeline. If the pipeline is cut off at entry level, who will be the senior technical and strategic leaders of the future?

It is often said that the jobs AI takes away will be replaced by more interesting and strategic roles further up the value chain: but will we have the talent to fill those roles if entry routes have been decimated lower down?

In our view, the time has come for corporates to review talent and hiring strategies, with key principles being:

  • Protect routes for junior talent: Maintain some level of junior hiring, including graduates, apprentices and career switchers. Nurtured well, they will become the lifeblood of your future business.
  • Upskill rather than replace: Rather than looking to eliminate headcount through AI, review where upskilling can enable your team to get more out of present tools. Empower your people to manage AI – instead of using AI to manage out people.
  • Prioritise diverse problem-solvers: A diverse intake of talent will provide a wider range of thinking and problem-solving capability. Organisations need people who possess the judgement to validate AI outputs and the curiosity to find the right business problems for AI to solve.
  • Keep training your tech teams on fundamentals: It is a challenge to know what training to provide tech teams given that things fall out of date so quickly. However, don’t abandon foundational computer science concepts, programming logic and algorithm design. Even if AI writes the code, humans need robust problem-solving and creative skills to know what to ask the AI to build and how to validate it. Keep a continuous learning culture and encourage the sharing of ideas and lessons learned.

Token-o-palypse is becoming a major concern for corporates. But it also an opportunity. At this early stage, there is a window for business leaders to ask themselves: Are we going to do something about this? The right decisions now could establish the blueprint for years to come.