Corporate America Adopts Model Routing as 95% of Enterprise AI Still Runs on Premium Models
Updated
Updated · CNBC · Jun 5
Corporate America Adopts Model Routing as 95% of Enterprise AI Still Runs on Premium Models
3 articles · Updated · CNBC · Jun 5
Summary
$200 in weekly token use per employee is pushing companies to route simple AI tasks to cheaper models instead of defaulting to frontier systems for everything.
Cisco's Jeetu Patel said that pace equals about $10,000 a year per worker and roughly $900 million annually for a 90,000-employee company, forcing budget reallocations after Cisco overshot its own plans.
Cognition CEO Scott Wu said routing can deliver 5 to 10 times better cost efficiency on routine work, while Glean estimates about 95% of enterprise AI usage still goes to the most expensive models.
That shift threatens OpenAI and Anthropic's premium-everywhere pricing model, leaving frontier labs with the hardest jobs while buyers gain leverage over how AI spending is allocated.
Cognition is also offering a productivity guarantee of up to $10 million in funded usage, underscoring a broader move to judge AI by measurable output and ROI rather than token volume.
How can businesses cut AI costs without sacrificing the breakthrough innovation promised by top models?
With free, state-backed Chinese models flooding the market, can American AI giants win the race to the bottom?
As corporations slash AI spending, are the sky-high valuations of labs like OpenAI built on a foundation of sand?
After 95% of GenAI Pilots Flopped, Model Routing Redefines Enterprise AI in 2026
Overview
By mid-2026, model routing has become a key standard for enterprise AI, driven by the growing complexity of AI tasks and the rise of many specialized models. Enterprises need AI systems that balance performance, cost, and flexibility to support digital transformation. However, integrating and optimizing these diverse AI capabilities is challenging. Model routing solves this by unifying different models into a single system, enabling the creation of general-purpose digital workers. This shift allows organizations to dynamically assign tasks to the best-suited models, making AI deployment more efficient and adaptable to changing business needs.