Across industries, companies are rapidly adopting artificial intelligence (AI) technologies to boost efficiency, but uncertainty remains around the true costs of scaling AI use. While businesses have traditionally controlled labor expenses, the costs associated with AI—measured in computing units called tokens—are often opaque and difficult to manage.
Founders and executives report that tracking token consumption per AI interaction is challenging, leaving many companies uncertain about their AI spending. Robert Stein, co-founder of Morvo, a Manchester-based startup that provides AI personal shoppers for e-commerce brands, noted that it is essentially impossible to determine exact token usage in conversations. To address this, Morvo has implemented its own token tracking system to better forecast client bills. Still, some clients confess a reluctance to assume full responsibility for the risks involved in AI deployment and cost control.
Similarly, Tom Vaughton, founder of Wiltshire-based marketing agency Varn, has developed a free AI tool for clients but fears that providers may increase costs by moving advanced features into higher-priced tiers, potentially creating a two-tier system where only wealthier companies can afford the most capable AI models. He raised questions about whether AI actually offers savings compared to human labor once the full costs are accounted for.
Despite widespread adoption—89 percent of firms reportedly use AI in some business function—only 46 percent have moved beyond pilot stages, indicating skepticism about AI’s productivity benefits. Studies show that while the unit prices of AI tokens have declined by about 50 percent due to improved model efficiency, overall enterprise spending can increase sharply. This phenomenon aligns with the Jevons paradox, where greater efficiency drives higher consumption.
Patrick Eden, CEO of Palma.ai, a San Francisco-based firm monitoring AI agent behaviour, emphasized the variability and opacity of token costs. Providers have subsidized AI pricing to build demand, but reliance on AI tools has grown so strong that outages can disrupt workflows significantly. Eden drew parallels between current AI market dynamics and early cloud computing, where low initial prices led to customer lock-in and eventual price hikes.
Financial data reflect the imbalance between industry expenditure and revenue: total AI spending is estimated at $1.9 trillion, while revenues amount to $972 billion. Leading companies like OpenAI remain unprofitable, and Meta recently warned it will require substantial capital—potentially hundreds of billions—to fund ongoing AI investments.
Not all industry leaders share concerns about vendor lock-in. Natasa Williams, CEO of London-based wealth management platform Cadro, believes competitive markets will allow businesses to switch AI providers as needed. However, some Cadro clients already restrict AI usage among employees due to cost worries.
Russell Brown, UK chief executive of technology provider SCC, urged companies to consider the economic value of AI on a per-task basis, rather than defaulting to the most powerful or expensive models. He analogized that firms should avoid “using an Aston Martin to deliver the post,” highlighting the need for cost-effective AI solutions tailored to specific tasks.
Infrastructure challenges may also drive AI costs higher, especially in markets with high energy prices. UK data centers face electricity costs approximately four times higher than those in the United States, a significant factor given AI’s heavy energy consumption. The trade body techUK warned that AI computing demand may soon exceed supply, potentially pushing prices upwards amid geopolitical uncertainties.
Finally, concerns about technological sovereignty persist. Tomás O’Leary, CEO of software company Origina, cautioned that businesses could repeat past mistakes by becoming overly dependent on a few dominant AI suppliers. Without careful planning, similar risks of vendor lock-in and reduced autonomy could arise in the AI landscape.