Champagne Consumption on Beer Budgets: The Emerging AI Risk Few Are Talking About

Images/Canva

AI costs money every time it reads or writes even a tiny piece of text.

As organisations rush to embed AI, often because everyone else seems to be doing the same thing and sometimes before proper audits are completed, one question few leaders are asking is this: given that current AI pricing models are unsustainable, what happens when the AI “happy hour” subsidies come to an end and we all have to start paying market rates for AI usage?

When AI Becomes a Co-Worker

This may come as a shock to organisations which, over the past months and years, have actively encouraged employees to use and embrace AI, in some cases even referring to AI agents as co-workers. Many organisations have embedded AI into workflows, tools and processes, and in some cases, replaced humans with AI as the customer’s first point of contact.

The Subscription Shock

Now imagine this scenario. Having embedded AI across your organisation, you receive an email from your AI vendor informing you that from next month your subscription will no longer be zero or a fixed monthly fee, but will instead be charged according to actual usage.

You think little of it. After all, your subscription has been £200 per month for some time. Then, at the end of the month, you’re presented with an AI bill running into three, four, or even five figures!

Highly unlikely?

When Even Uber Gets Caught Out

According to Forbes (17 May 2026), Uber burnt through its entire 2026 AI budget in just four months. This was driven primarily by more than 5,000 developers relying on AI for 70% of their code, forcing the company to rethink its AI consumption governance. If a company that is expert in surge-charging can be caught out by AI (token*) pricing models, what chance do other organisations have? 

Who Is Watching the AI Meter?

So, who took their eye off the ball? Was it the Board, the CEO, the CFO or the CTO? Was no one watching the token pump? And, going forward, whose job will it be to monitor the AI meter, and when do they press the kill switch?

“But I’m Not a Coder”

“I’m not a coder,” I hear you say. “I only use AI for small, repetitive tasks so why should I be concerned?”

Well, because you’re probably relying on large language models (LLMs), such as ChatGPT or Claude. These are vast machines and means even a seemingly small task requires significant resources. Imagine if summarising a 200-page PDF costs you £20 when subsidies disappear? Now multiply this by hundreds of employees.

From SaaS Certainty to Usage Volatility

So yes, you too will be impacted when AI subsidies come to an end because the current (AI) pricing models are just not sustainable. 

When organisations start paying market rates, unlike traditional SaaS, AI services will most likely be billed by usage, typically token consumption, rather than a predictable per-seat licence. This is assuming most of your LLM requirements are in English rather than global, multi-lingual environments, where risks could be amplified. For many LLMs, tasks involving translation or complex language structures consume more tokens, increasing costs, and can result in nuance or local context being diluted in the process.

The Economics No One Wants to Talk About

The reality is that frontier AI models cost far more to run than they are currently generating in revenue. The vast majority of ChatGPT users, for example, are on freemium, with only ~5% paying subscribers. It’s no secret that OpenAI is burning cash and according to Forbes (2 April 2026), the company is projected to make a loss of ~US$ 14 billion in 2026. 

Even when revenue figures are reported, it is difficult to assess whether these numbers reflect the full picture, given that circular AI funding muddies the water i.e. Company A invests in Company B, enabling company B to buy compute from Company A, and some of these (circular) transactions may be recorded as revenue. 

When the Music Stops

This is little more than a game of musical chairs and, as we know, at some point, the music stops. When it does, who will be left running around looking for a non-existent empty chair?

The £300,000 Question

Now imagine in the near future, being landed with an AI bill for £300,000 because AI was tasked with completing a project that two humans could have delivered for £70,000. How do you move forward? 

Do you now need to unpick and map AI usage across your organisation? If so, what stays, what goes, and how long will that take?

Are We Using a Sledgehammer for a Nut?

Do organisations really need to use giant LLMs for simple tasks such as summarising emails, resetting passwords or tracking orders?

This is akin to using an industrial mega-washer, as found in laundrettes, to wash two pairs of socks and then using it again to wash the small towel you forgot to include in the first round.

The Questions Leaders Must Ask Now

Before embedding AI too deeply, leaders need to ask themselves:

1. What problem(s) are we trying to solve? Start with a low-value, low-risk task.

2. Do we actually need AI for this, or could automation, process improvement or staff training and/or redeployment suffice?

3. If we do need AI, do these tasks really need LLMs (big models which are slower and more expensive to run), or could SLMs (Small Language Models), that respond faster, be sufficient?

4. Furthermore, if LLMs are required for some use cases, could SLMs handle the majority, MLMs (Medium Language Models) support others, and LLMs be reserved for perhaps the top 5% of tasks? 

The Real AI Risk

For organisations, the real AI risk will not only be hallucinations or output quality. It will be the realisation that their AI strategy has been built on unsustainable pricing models they have not yet learned to manage.

Perhaps once we all start paying true market price for LLM-driven actions, this will focus minds on just how wasteful current AI usage can be.

From Happy Hour to Hangover

This is why leaders must consider whether they can deliver intelligence at scale efficiently and, crucially, affordably.  It is also why going forward, teams responsible for AI adoption must work closely with those managing AI spend. 

Otherwise, organisations will continue ordering drinks assuming they're still on happy-hour rates, only for reality to become painfully clear the following morning, along with the inevitable hangover.

* A token is a tiny chunk of text, such as a word (dog), a punctuation mark (“.” or “!”), or even part of a word (prefix/suffix).

Sources: ExecutiveGlobalCoaching.com, Anand Logani (Forbes, 17 May 2026), Sandy Carter (Forbes, 2 April 2026)

 *****************************

Click HERE to learn more about me. Subscribe below to receive an alert as soon as I publish new editions of my LinkedIn (monthly) newsletters or, to read previous editions:

1. Responsible AI - (this one) Putting People and Culture at the heart of AI Strategy.

2. Leading with Emotional Intelligence (EQ) - Using EQ to build a culture of empathy and collaboration to drive organisational success

3. Inclusive Leadership in the era of AI - Creating a culture in which everyone thrives

#ExecutiveLeadership #AILeadership #ResponsibleAI #FutureOfWork #DigitalStrategy #CLevelInsights #AIGovernance