The hidden cost of AI adoption: Why enterprises are worried about token economics

For the last two years, artificial intelligence has largely been sold as a productivity revolution. Companies across sectors rushed to integrate AI tools into coding, customer support, analytics, content generation, and daily workflows, convinced that automation would eventually lower costs and improve efficiency.

But as enterprises scale up AI usage internally, a new concern is quietly emerging behind the hype, the economics of using AI may be far more expensive than many initially expected.

The debate gained attention recently after reports and discussions online highlighted how some large technology companies are beginning to closely monitor employee AI usage due to rapidly rising compute and token-related costs. The larger concern is no longer just about building powerful AI systems, but whether companies can afford to run them sustainably at scale.

WHY AI BILLS ARE RISING SO QUICKLY

Prabind Singh, Founder & Managing Director, Europa Technosoft Pvt Ltd, said that companies initially underestimated the infrastructure costs attached to enterprise-scale AI adoption.

“AI does improve efficiency, but many companies underestimated the infrastructure cost behind large-scale AI adoption,” he said.

He explained that continuously running AI systems requires enormous computing power, cloud infrastructure, GPUs, storage capacity, and API usage. While AI pilots may appear manageable initially, costs can rise sharply once adoption spreads across departments and workflows.

“Initially, companies experimented with AI in limited workflows, but once usage expanded across departments, the operational costs increased exponentially,” he added.

WHAT EXACTLY ARE TOKENS — AND WHY DO THEY MATTER?

At the centre of this issue lies something many ordinary users rarely think about: tokens.

“Tokens are essentially the units of text that AI models process while reading or generating responses,” Prabind explained.

Every AI prompt, uploaded document, code request, report summary, or chatbot response consumes tokens. Individually, the cost per interaction may appear small. But inside large companies with thousands of employees using AI tools all day, those costs can escalate rapidly.

“For example, if thousands of employees begin using AI tools throughout the day for emails, reports, coding, analytics, customer support, or document processing, the cumulative token usage becomes enormous,” he said.

Since most AI companies charge enterprises based on usage, the economics begin changing dramatically at scale.

AI IS STARTING TO LOOK LIKE INFRASTRUCTURE

One major shift companies are now realizing is that AI behaves less like ordinary software and more like a long-term infrastructure investment.

“Yes, many companies initially viewed AI as a productivity layer rather than as a continuously running infrastructure system,” Prabind said.

Once AI becomes embedded into daily operations, employee dependence on these systems grows rapidly and becomes difficult to scale down.

“At scale, organisations must manage not only model costs, but also cloud infrastructure, cybersecurity, compliance, data storage, integration layers, and governance frameworks,” he said.

That means enterprises are now dealing with multiple hidden costs beyond just paying for AI models themselves.

ARE COMPANIES ACTUALLY SAVING MONEY YET?

Despite the productivity gains, many firms are still struggling to see clear financial returns from AI.

“In most cases today, companies are not yet seeing net savings because they are still in the experimentation and integration phase,” Prabind said.

He noted that while AI is helping employees work faster, the cost of integration, employee training, infrastructure upgrades, and workflow changes is still very high.

“Productivity gains are visible, but financial ROI often lags due to high integration, training, and change management costs,” he added.

This is becoming one of the biggest questions facing enterprise AI adoption today: even if AI improves output, are companies truly saving money yet?

THE REAL CHALLENGE IS NO LONGER BUILDING AI

Interestingly, experts now believe the bigger challenge may no longer be building powerful AI models.

“Today, the bigger challenge is increasingly cost control and sustainable scalability rather than simply building AI models,” Prabind said.

As AI tools become more powerful and accessible, enterprises are shifting focus toward optimisation — deciding which models to use, controlling token consumption, managing infrastructure efficiently, and balancing performance against cost.

“In many ways, the AI industry is now entering a phase similar to early cloud computing, where scalability and cost optimisation become just as important as innovation itself,” he said.

WILL AI BECOME TOO EXPENSIVE FOR SMALLER FIRMS?

The concern is especially important for smaller businesses that may not have the financial resources of Big Tech firms.

“High AI costs could definitely slow adoption for smaller businesses if the ecosystem remains dependent on expensive centralised infrastructure,” Prabind said.

However, he believes the market is already adapting through smaller specialised models, open-source AI systems, and hybrid deployment strategies that could gradually reduce costs over time.

THE AI COST PROBLEM MAY NOT LAST FOREVER

Despite current concerns, experts do not believe the economics of AI are permanently broken.

“Yes, I believe AI costs will reduce significantly over time, although demand for computing power will continue to grow,” Prabind said.

He compared the current phase to earlier technology revolutions like cloud computing and internet infrastructure, where costs initially appeared unsustainable before eventually becoming more affordable and efficient.

Over the next few years, advances in AI hardware, model efficiency, open-source ecosystems, and localized deployments are expected to reduce costs significantly.

For now, however, the excitement around AI is increasingly being matched by a tougher business question: not just what AI can do, but how much companies are willing to spend to keep using it at scale.