When ChatGPT launched in late 2022, the pitch to businesses was simple: AI would make work faster and dramatically cheaper. Fast forward to 2026, and for a growing number of enterprises, that's no longer true.
Companies acted on that early promise fast - across tech and service industries, roles were cut on the assumption that AI could do the work of ten people for a fraction of the cost. Today, the financial picture looks very different. In some cases, AI is now the more expensive option.
How We Got Here: "Token Maxing"
Part of the problem is cultural, not just technical. Through 2025, heavy AI token usage became something of a status symbol inside parts of the tech industry - engineers were encouraged, and in some workplaces effectively required, to route even trivial tasks through AI models just to be seen as "keeping up." The result was a wave of artificially inflated usage: people running up token counts to look productive on internal metrics, rather than because the task genuinely needed it. At some firms, individual engineers were racking up hundreds of thousands of dollars a year in token spend - not because the work demanded it, but because the incentives did.
The Bill Has Landed
That inflated demand collided with a separate problem: global shortages in the data centre components needed to run these models at scale. Together, the two forces have pushed prices up sharply. Since late 2025, average LLM token pricing has more than doubled - from roughly $0.11 per million tokens to around $0.22 by mid-2026.
At enterprise scale, that shift is brutal:
- AT&T is reportedly processing close to 8 billion tokens a day.
- Meta burned through an estimated $900 million in tokens in a single month.
- Microsoft and Uber have already started pulling back or cancelling licences for higher-cost tools, with some executives describing the spend as "like driving a Ferrari" - impressive, but hard to justify for the daily commute.
Tokens vs. Talent: The Numbers Are Shifting
The result is a genuine reversal for certain categories of work. AI is still clearly the cheaper option for complex coding and similarly high-value tasks. But for things like data entry or call-centre style work, running AI agents at scale is now often comparable to - or more expensive than - simply employing a person to do it.
Finance teams are increasingly facing a question nobody expected to be asking three years ago: tokens, or humans? With around three-quarters of executives expecting their technology budgets to rise, and AI already accounting for roughly a fifth of enterprise tech spend, the era of AI feeling essentially free is over.
What This Means Going Forward
There's little reason to expect this to settle down soon. As the major AI labs move toward IPOs, they'll face real shareholder pressure to turn a profit - which almost certainly means further price rises for tokens and AI agents, not fewer. For a lot of businesses, the tool they brought in to replace their workforce may end up costing more than the people it replaced.
The Takeaway for Anyone Running Analytics or Automation Projects
This isn't an argument against AI. It's an argument for the same discipline that should apply to any significant technology investment: model the real cost, not the sticker price, before you commit.
A few practical questions worth asking before scaling any AI-driven workflow:
- What's the actual per-task cost at your expected volume - not the demo cost, the production cost, run over a full month?
- Is usage being driven by genuine need, or by internal incentives - dashboards, leaderboards, "look how much we're using AI" reporting - that quietly inflate token spend?
- Have you priced the human-labour alternative properly, including the parts AI doesn't remove - oversight, error correction, exception handling?
- What happens to your unit economics if token prices double again in the next 12 months, as they already have in the last?
These are exactly the kinds of questions that come up when I run a Power BI health check or a diagnostic review for a client - not "is this technology impressive," but "does the maths actually hold up once it's running at real volume." AI, automation, and BI tooling all deserve the same scrutiny you'd apply to any other line item on a budget. The businesses that come out ahead over the next few years won't be the ones that adopted AI fastest - they'll be the ones that measured it properly.
Evaluating where AI or automation genuinely earns its place in your reporting and operations - versus where it's quietly costing more than it saves?
Book a Free 30-Min Scoping Session