The latest
Corporate America is changing how it spends on artificial intelligence. After a period when employees were rewarded for consuming more tokens, companies are now selecting the best model for each task instead of using the most powerful and expensive system for everything.
The new strategy uses frontier models for planning and complex problems, then assigns execution and review to cheaper or open-weight systems. A Cursor experiment found that building a browser entirely with one OpenAI model cost more than $10,000, compared with $1,339 when combining Cursor’s model with one from Anthropic.
Chinese models such as Kimi, DeepSeek and GLM are benefiting from the shift, while U.S. labs are offering free usage, incentives and partnerships to retain customers. Loyalty to a single provider is becoming rare as model performance changes several times a week.
Details
- Price war: Customer-support platform Pylon received an estimated $1.6 million in free usage from one provider this year.
- Chinese adoption: About half of Hex’s customers added Moonshot’s Kimi to their workflows within two weeks.
- Corporate savings: Zoom uses a combination of models from Meta, Anthropic and OpenAI to reduce costs.
- Alternative cost: Telnyx estimated that continuing with Anthropic under usage-based pricing would cost about $100,000 a day.
- Task division: Companies are using a powerful model to direct work, open models to execute it and another system to review the results.
- Regulatory debate: Nvidia, Microsoft and Palantir backed open models and cautioned policymakers against imposing hasty restrictions.
Between the lines
Companies are not retreating from AI. They are abandoning the idea that it must be purchased from one provider at any price. Models are becoming interchangeable commodities, while value shifts toward platforms that can select, combine and manage them efficiently.
What to watch
The impact will become visible in the pricing of closed models and the ability of Chinese providers to expand in the U.S. market. If companies continue diversifying suppliers, leading labs may have to cut prices and offer simpler models instead of relying only on premium frontier systems.
