For the past two years, much of the conversation around AI has focused on capability. Which model is smartest? Which coding agent is fastest? Which AI assistant saves the most time?
But according to new research from Gartner, enterprises may soon face a very different question:
Can they afford to keep using AI at scale?
Gartner predicts that by 2028, the cost of AI token consumption for software developers could meet or even exceed the average developer’s salary. As organizations rapidly deploy coding agents, autonomous workflows, and increasingly sophisticated AI systems, token usage is emerging as one of the largest—and least understood—cost centers in enterprise technology.
The Shift From Software Licenses to Consumption Economics
The traditional software model was simple. Companies purchased software licenses on a per-user or per-seat basis, making costs relatively predictable.
AI is changing that model entirely. Rather than paying a fixed monthly subscription, enterprises are increasingly paying based on consumption: every prompt, every context window, every generated line of code, and every agent interaction consumes tokens. As AI models become more powerful and agentic workflows become more autonomous, organizations are discovering that intelligence itself is becoming a metered resource.
According to Gartner senior principal analyst Nitish Tyagi:
“Organizations are rapidly moving from experimentation to scaled deployment of AI coding agents, but many are underestimating the financial impact of rising token consumption.”
The result is a new operating reality where software budgets increasingly resemble cloud infrastructure bills.
The Numbers Are Already Starting to Become Uncomfortable
While Gartner’s prediction is based on a global average developer salary of approximately $2,000 per month, the underlying trend extends far beyond any single geography.
Tyagi reports hearing examples that would have seemed unimaginable only a year ago:
“I have heard scary numbers like ‘My developer consumed $20K last month,’ or ‘A business user consumed $32K.'”
These figures are not necessarily representative of average usage. But they illustrate an emerging challenge facing enterprises deploying frontier AI systems at scale: costs can escalate rapidly when organizations lack visibility into how AI agents consume resources.
This is particularly true for autonomous agents, where humans are no longer directly controlling every interaction. A single agent running continuously across multiple workflows, repositories, or business systems can consume enormous amounts of compute without anyone noticing until the invoice arrives.
Why AI Costs Are Rising Faster Than Expected
Part of the challenge lies in how AI systems themselves operate. Large context windows, multi-agent orchestration, repeated reasoning cycles, autonomous verification loops, and continuous background processing all consume tokens. As organizations move beyond simple chat interfaces and begin deploying AI agents that can work for hours or days autonomously, token consumption can increase exponentially.
At the same time, AI vendors themselves face growing infrastructure costs. Training larger models, building inference infrastructure, and maintaining profitability are pushing the industry toward increasingly sophisticated consumption-based pricing models. Unlike traditional SaaS software, where costs remained relatively stable regardless of usage patterns, AI introduces variable costs that fluctuate based on behavior.
According to Gartner, many organizations currently lack both the visibility and the operational maturity required to effectively manage these new cost structures.
More Tokens Does Not Mean More Productivity
One of Gartner’s most important findings challenges a common assumption in enterprise AI adoption: that greater AI usage automatically translates into greater productivity.
According to Tyagi:
“Tokenmaxxing is not directly related to higher productivity gains, but optimizing token consumption is.”
This observation represents a fundamental shift in how organizations should think about AI deployment.
The goal is not to maximize AI usage. The goal is to maximize business outcomes while minimizing unnecessary computational overhead. In many cases, carefully engineered prompts, optimized context windows, and intelligent model selection produce better results than simply increasing token consumption.
In other words, efficiency may become a more valuable skill than access to the largest models.
Context Engineering Is Becoming a Business Skill
As enterprises mature their AI operations, a new discipline is emerging: context engineering.
Rather than asking how to get AI systems to do more, context engineering asks how to provide AI systems with precisely the information they need—and no more.
Gartner recommends several approaches:
Organizations should classify tasks into different levels of AI autonomy, ranging from developer-led workflows to fully autonomous agent systems. Smaller models should be used whenever possible, with frontier models reserved for high-value or highly complex tasks. Developers should be trained to optimize context windows, summarize information efficiently, and eliminate unnecessary token consumption.
Perhaps most importantly, organizations need governance systems capable of monitoring, auditing, and controlling AI usage at scale.
Without these controls, AI spending can quickly outpace the productivity gains it was intended to create.
The Future Competitive Advantage May Be Cost Discipline
The first phase of the AI revolution was about access.
The second phase has been about capability.
The third phase may ultimately be about efficiency.
As AI models become increasingly commoditized and widely available, competitive advantage may shift toward organizations that can deploy intelligence most effectively while maintaining economic discipline. Companies that treat AI as an unlimited resource risk finding themselves with growing bills and diminishing returns.
Those that master context engineering, workflow optimization, and AI governance may ultimately gain the greatest advantage.
The future of enterprise AI may not belong to the companies that consume the most intelligence.
It may belong to the companies that consume just enough.
Sources: Gartner Research, AI Coding Costs to Surpass Average Developer Salary by 2028 (June 2026); InfoWorld reporting on enterprise AI token consumption trends.
