As the number of tokens consumed by artificial intelligence (AI) agents and applications continues to climb, organizations are paying much closer attention to cost. It feels like the conversation has shifted almost overnight. Not long ago, the focus was on tokenmaxing—maximizing AI use wherever possible. Now, the emphasis is moving toward tools, platforms, and services that can reduce token consumption and make AI more economical to operate.
The rising cost of AI
That shift is happening for a reason. Organizations are discovering that it is much easier to increase AI usage than it is to measure whether that spending delivers meaningful business value.
Uber recently offered a high-profile example. The COO of Uber recently noted that despite massive adoption of automated engineering tools, it remains difficult to connect those costs to meaningful customer outcomes. That revelation came shortly after the CTO for Uber disclosed that the company had exhausted its entire 2026 AI coding tools budget in just four months.
That kind of sticker shock is becoming harder to ignore. As AI adoption scales, many organizations are realizing they need a more disciplined approach to controlling usage and aligning spend with outcomes.
Why costs are coming under scrutiny now
Concerns about AI costs are rising just as providers push toward IPOs. To attract investors, they’re emphasizing paths to profitability. For years, many providers subsidized AI services, but mounting losses are driving them to pass those costs—plus margin—on to customers.
That means customers should expect more of those costs to get passed through—along with a healthier profit margin. In other words, the era of artificially cheap AI may be coming to an end.
How organizations are starting to respond
As costs rise, organizations are becoming more deliberate about how they deploy AI.
One major area of focus is reducing the amount of data included in a prompt’s context window. A growing number of tools and platforms—from serverless computing frameworks and databases to graphs and indexes—are designed to make frequently used data accessible without forcing it into every prompt. The less data an AI model needs to process, the fewer tokens it consumes.
At the same time, organizations are becoming more selective about which models they use. Not every task requires access to the latest frontier model, which is usually also the most expensive. In many cases, routine tasks can be handled just as effectively by older, less expensive models. That kind of model discipline can significantly lower costs without sacrificing outcomes.
Infrastructure choices are also starting to matter more. Alternatives to graphical processing units (GPUs) are beginning to gain traction, especially for AI inference engines running in production environments. As these options mature and become more widely available, they should help lower the overall cost of delivering AI services.
A growing opportunity for MSPs
All of this creates a significant opportunity for managed service providers (MSPs).
As AI cost optimization becomes a priority, organizations will need partners that understand how to manage usage, select the right models, reduce token consumption, and architect services in a more efficient way. In many respects, AI cost management is becoming the next major area of operational expertise.
The challenge for MSPs is familiar: gaining those skills before they become commonplace. Providers that invest early in the expertise needed to optimize AI environments will be in a strong position to build differentiated managed AI services. Those that wait may find themselves competing in a crowded market where cost optimization has already become table stakes.
Photo: CeltStudio / Shutterstock
