Executive Summary
The May 2026 AI Gateway Production Index reveals a significant bifurcation in the AI model market. New, low-cost models like DeepSeek are capturing massive token volume for high-throughput tasks, while expensive, frontier models from labs like Anthropic are solidifying their dominance in high-stakes applications, driving overall spend up. This data indicates a maturing market where enterprises are implementing sophisticated routing strategies to optimize for both cost and quality, selecting the right model for the right job rather than relying on a single provider.
Key Takeaways
* Market Bifurcation: The AI market is clearly splitting into a value tier for high-volume tasks and a premium tier for high-stakes work. Overall token usage grew 20% month-over-month, while total spend grew much faster at 43%.
* DeepSeek's Explosive Growth: Newcomer DeepSeek saw its token share surge from less than 1% to 17% in one month, becoming the third-largest provider by volume. However, its share of total spend remained near 1%, highlighting its extremely low price point.
* Anthropic's Spend Dominance: Anthropic's share of total spend increased from 61% to 65%, commanding 70-80% of the budget for critical use cases like AI app generation, back-office agents, and coding agents.
* Smarter Routing is a Key Strategy: Companies are becoming more deliberate about cost management. They are routing high-volume, lower-risk work to cheap models like DeepSeek V4 Flash while reserving expensive frontier models for tasks where quality is non-negotiable.
* Cost-Consciousness in Action: The slow adoption of Google's newer, more expensive Gemini 3.5 Flash model (7% of Flash family tokens) compared to the established Gemini 3.0 Flash (90%) demonstrates that teams will not upgrade if the cost-benefit isn't clear.
Strategic Importance
This report signals a major shift in the enterprise AI landscape, moving from a "best model wins all" mentality to a sophisticated, cost-aware portfolio approach. The ability to effectively route requests to different models based on cost and capability is becoming a critical competitive advantage for businesses building with AI.