Smart Spending: Why Companies Are Rethinking Their AI Model Strategy

Corporate finance chiefs are taking a harder look at AI costs, and the shift is starting to reshape how businesses buy and use these tools. For the past two years, the standard approach was simple: route every query through the most powerful model available, regardless of the task. But with AI expenses blowing past budgets, companies are asking whether every job really needs a top-tier model.

Enter model routing—a straightforward fix that matches each task to the right tool. Hard problems go to expensive frontier models; routine work goes to cheaper, faster alternatives. Scott Wu, CEO of AI coding agent Devin, says the savings on boilerplate tasks can be five to ten times better. Yet most companies aren't routing at all. Glean CEO Arvind Jain estimates that 95% of enterprise AI usage still runs on premium models, even for simple queries like naming the third U.S. president. (It was Thomas Jefferson.)

The math is sobering. Cisco's chief product officer Jeetu Patel notes that at roughly $200 per employee per week in token usage, a 90,000-person company faces a $900 million annual bill. Cisco itself came in well over budget and has reallocated resources, prioritizing tokens over other spending.

This cost pressure is forcing a change. Cognition now offers an AI productivity guarantee: if Devin delivers less engineering value than the customer pays for, Cognition will fund usage up to $10 million to make up the difference. The idea is to measure output—actual human hours saved—not just tokens consumed.

For OpenAI and Anthropic, the implications are significant. If companies steer easy, high-volume work to cheaper open-source models, those labs only get the complex, high-margin jobs. Patel doesn't think that kills the frontier labs, but he predicts a pricing model shift: labs will need to get more efficient rather than simply charging more. The bottom line: pricing power is moving from AI sellers to AI buyers.