A four-person startup called Swan AI is paying $113,000 a month for AI. Four people, six figures in monthly compute bills. That’s not a staffing problem. That’s a pricing problem.
By mid-2026, the economics of running AI workloads have gotten so brutal that companies of every size are doing something the big model providers really don’t want them to do: shopping around in real time. Businesses are now dynamically routing queries through multiple AI models, picking the cheapest option that can handle each specific task.
The budget crisis nobody planned for
The numbers tell a story that should make any CFO uncomfortable. Corporate AI investments hit $252.3B in 2024, a staggering sum that suggested the industry was betting big on transformative returns. The problem? Most companies adopting AI report cost savings of less than 10%.
And the operational side is even uglier. Uber’s CTO revealed that the company burned through its entire 2026 AI budget by the second quarter. Not a partial overrun, not a slight miscalculation. The whole budget, gone before summer.
Swan AI’s $113,000 monthly tab is particularly striking given the startup’s size. For a team of four, that’s roughly $28,250 per person per month in AI costs alone, before salaries, office space, or anything else that keeps a business running.
The great model shuffle
Mid-2026 data points to a surge in enterprise adoption of AI routing and orchestration tools that automatically switch between models based on cost and capability. Not every query needs the most powerful model available. A simple text classification task doesn’t require the same firepower as generating a complex legal analysis.
The competitive pressure from Chinese AI models has turbocharged this trend. ByteDance was offering AI services priced as much as 99.8% below GPT-4 rates back in 2024. DeepSeek’s lower-compute models, which arrived in early 2025, further accelerated the cost-efficiency movement.
OpenAI and Anthropic feel the squeeze
OpenAI is reportedly weighing significant reductions in its enterprise token pricing. Anthropic is expected to follow.
For investors tracking this space, the margin compression implications are significant. If the two most prominent Western AI model providers are forced into a price war with Chinese competitors and open-source alternatives simultaneously, the path to profitability gets considerably longer. The $252.3B that poured into corporate AI in 2024 was predicated on assumptions about pricing power that are looking increasingly fragile.