Tokenomics: Bain & Co Warns AI Costs Outpace Revenue

What happened

Bain & Company has published analysis warning that the economics underpinning generative AI are increasingly strained, with both vendors and enterprise users caught in the same cost spiral. The consultancy's "tokenomics" framing argues that AI providers' infrastructure and compute costs are rising faster than the revenue they generate, while the businesses buying AI services are simultaneously seeing their own token-based usage bills climb in ways that are hard to predict or control.

According to the analysis, these two problems are not separate — they are structurally linked. The same dynamics that make it expensive for vendors to run large language models at scale are the dynamics that make consumption-based pricing unpredictable for the organisations paying for tokens. As usage grows, so do the underlying compute demands, and neither side of the transaction has yet found a durable way to bring costs and value back into balance.

Why it matters

This is fundamentally a story about the sustainability of the AI business model, not a feature-level update. Token-based pricing was adopted industry-wide because it felt fair and transparent — pay for what you use — but Bain's analysis suggests it has instead created a mismatch where usage scales unpredictably against value delivered. For technology and transformation leaders, this raises a practical governance question: AI consumption needs the same budget discipline, forecasting and monitoring that organisations apply to cloud spend, rather than being treated as a flat software licence.

For vendors, the implication is sharper still. If the cost of serving each additional token continues to outpace what customers are willing or able to pay, pricing models built purely on consumption may need to evolve — toward bundling, outcome-based pricing, or tighter efficiency gains in the underlying models themselves.

The Renascence take

The tokenomics gap is a textbook case of a pricing model outrunning the behavioural reality of how people actually use a service. Consumption-based pricing assumes rational, visible trade-offs at the point of use — but most AI users have no real-time sense of what a given prompt, query or workflow is costing until the invoice arrives.

Token-based pricing fails the same way many "pay for what you use" models fail: it externalises the cost-awareness problem onto the user at the exact moment they're least equipped to manage it. The fix isn't just better vendor pricing — it's giving buyers real-time cost visibility and defaults that nudge efficient usage, the same behavioural-design discipline that made cloud-cost governance mature a decade ago. Any organisation scaling AI adoption without that visibility is effectively flying blind on its own margins.