Enterprise leaders are experiencing sticker shock over rising AI costs, with a KPMG survey finding that 29 percent of senior executives struggle to understand scaling costs and nearly half are reconsidering their deployments. Gartner research warns that AI agent costs for coding could exceed developer salaries by 2028 on a global average basis, and are already higher than salaries in lower-wage regions like India. The shift by major providers to usage-based, per-token billing—away from flat-fee subscriptions—has made costs much less predictable and harder to justify.
Summaries like this, in your inbox every morning.
Sign up free →What happened
A KPMG survey of over 2,000 senior executives across 20 countries found that 29 percent struggled to understand operating costs as enterprise AI deployments scale, and nearly half were reconsidering their AI rollouts when costs exceeded expected value. Model providers including Anthropic, OpenAI, and GitHub have shifted from flat-fee subscription models to usage-based billing charged per token.
Why it matters
Gartner research suggests that in some regions, the cost of AI coding agents is already exceeding developer salaries, and by 2028 the average cost per developer globally could surpass typical salary levels. This makes the economics of AI deployment unsustainable for many organizations, forcing them to rethink whether heavy usage actually delivers proportional productivity gains—a core justification for the technology investment.
What to watch
Executives are exploring lower-cost and smaller models as alternatives to expensive flagship systems, and some are optimizing token usage rather than maximizing it. Industry capex for AI datacenters is projected at $1.5 trillion（約240兆円） over five years until 2030, meaning costs will eventually have to be recouped from customers, intensifying pressure on pricing models.
Enterprise organizations rushed into AI adoption following the trend, but are now confronting an unfamiliar cost structure. The shift from subscription-based pricing to usage-based, per-token billing by major model providers created unpredictability in expense management—a shock for finance and operations teams accustomed to cloud pricing models with clear optimization tools and transparency. KPMG's finding that 29 percent of senior executives struggle to understand scaling costs suggests the industry has outpaced corporate procurement maturity.
The Gartner research on AI-assisted coding reveals the deeper economic problem: there is no proven link between increased token consumption and productivity gains. In fact, careful, controlled usage appears to yield both lower costs and higher code quality. This decouples the primary revenue driver (token consumption) from the value proposition, creating a fundamental mismatch. When agent costs are already exceeding developer salaries in India and are on track to exceed global average salaries by 2028, the technology becomes a cost center rather than a multiplier—a reversal of the original business case.
The $1.5 trillion（約240兆円） capex forecast for AI datacenters through 2030 underscores that these costs must eventually be recovered from customers. Vendors face a narrowing path: charge too little and remain unprofitable; charge too much and customers shift to cheaper open-source alternatives or redeploy with smaller, lower-cost models. The absence of vendor billing standardization gives enterprises additional negotiating leverage and incentive to optimize internally rather than maximize external usage.
No discussion yet for this article
Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime
1 minute a day. The AI essentials.
200+ sources · Email / LINE / Slack
