For a brief window, companies had access to the most transformative technology in a generation at the cost of a streaming subscription. Tools like ChatGPT put AI within reach of anyone with a browser and time for experimentation, while GitHub Copilot came in at just $10 a month, with token costs remaining relatively low. In the beginning, experimentation felt cost-effective, easy and relatively low-risk.
But that era is ending, and the bill is coming due faster than a lot of enterprise leaders anticipated.
For example, GitHub is moving all Copilot plans to usage-based billing, where every interaction consumes tokens. This includes input, output and cached tokens, priced at published API rates. What used to be a predictable monthly seat fee is now a variable cost that scales with how intensely engineers use Copilot. When you run out of credits, you stop working.
Why Are Enterprise AI Costs Increasing?
Enterprise AI costs are rising due to a shift from flat-rate subscription models to usage-based billing. Under this consumption model, every prompt, iterative interaction and multi-agent workflow incurs direct token costs. Additionally, newer agentic AI systems consume significantly more tokens per task than standard chat tools, outpacing falling unit prices and creating unexpected budget exposure.
The AI Meter Is Running
For decades, software licensing followed a simple model: Companies paid a flat fee per seat, and teams used tools as needed. A junior analyst who spends hours wrestling with Excel formulas costs a company no more in licensing fees than an expert who completes the same task in minutes.
AI has disrupted that model. Now, every misfired prompt, vague instruction and iterative exchange with an agent carries a direct cost. And so do multi-agent workflows that complete tasks with little or no human intervention. In both cases, the meter is always running.
Uber offers a recent example of how quickly this shift can surface. After deploying Claude Code to roughly 5,000 engineers in December 2025, the company exhausted its entire 2026 AI tools budget within four months.
COO Andrew Macdonald noted the difficulty of linking increased AI usage to measurable product improvements. Internal incentives, including leaderboards ranking teams by AI usage, drove rapid adoption. The outcome was predictable: usage surged, costs escalated and return on investment is unlikely to be fully measured.
This introduces a new form of cost exposure that many finance and engineering leaders still have to fully quantify. A team of 20 engineers with uneven AI fluency is not just less efficient than a highly fluent team; it is more expensive. With API costs ranging from $500 to $2,000 per engineer per month, inefficiency translates directly into higher spending.
At the individual level, AI fluency is fast becoming a form of professional leverage. Those who combine it with strong analytical capacity are the talent companies compete hardest to hire and retain.
Even as model costs decline, overall enterprise spending may continue to rise. Gartner projects that inference costs could fall by as much as 90 percent by 2030, but total AI bills are still expected to increase. Newer, agentic systems require significantly more tokens per task than the chat systems that have been around for several years, and rising consumption can easily outpace falling unit costs.
The Economics of AI
The AI ecosystem is also constantly shifting. AI providers are under pressure to align revenue with the massive infrastructure investments already made. Microsoft, Google and Amazon have collectively spent billions expanding AI capacity.
Providers lose money on some subscription plans, where heavy usage exceeds the value of the monthly fees. As the market strives to find balance and sustainability, greater pricing discipline for vendors and greater spending discipline for customers are both becoming unavoidable.
This creates tension for businesses that formed expectations during the era of inexpensive AI access. For example, customer service automation becomes more complex when inference costs scale with every interaction. And even modest, distributed usage across large teams can accumulate into a meaningful budget line.
Studies show that, in certain workflows, AI systems can prove more expensive than human labor once oversight, corrections, infrastructure and failed outputs are factored in. This sheds new light on the assumption that AI universally reduces costs. Sometimes, it doesn’t.
In this new economics, obsessing over token-level optimization is not a viable strategy. The solution is determining the right level of AI spending for each business. That will look very different for a 20-person startup than for an enterprise with 5,000 engineers. Getting it right requires access to metrics most organizations are still trying to figure out, such as error rates and the frequency of human rework. Organizations need granular, real-time visibility into where AI delivers real value to succeed in this ever-changing AI landscape. To get this visibility, companies need a centralized view that allows them to understand AI usage patterns, measure adoption and proficiency, identify potential risks and assess business impact and ROI.
AI Habits Meet New Cost Realities
For enterprise leaders, the challenge extends beyond cost control. The habits formed during the period of low-cost AI are turning into rapidly growing expense items that companies didn’t have a few months ago. Teams once rewarded for maximizing usage may be expected to justify tokens consumed.
In this reality, measurement is key. So is the optimization of spending. Traditional metrics such as usage volume, AI-assisted commits or hours saved were sufficient under flat-rate pricing models. They fall short in a consumption-based environment where budgets are finite and costs scale with activity.
AI will ultimately need to demonstrate that it is both cost-effective and predictable at scale. That threshold has not yet been fully met. In the meantime, organizations are beginning to recognize that the era of effectively free AI experimentation is coming to a close.
