Every month, AI Gateway routes tens of trillions of tokens between production applications and AI labs, giving us visibility into what AI usage actually looks like, separate from leaderboards and benchmarks.
May 2026 production index summary
Total AI Gateway tokens grew +20% MoM; total spend grew +43% MoM. Customers paid almost 20% more per token on average than in April.
DeepSeek’s share of tokens jumped from under 1% to 17% in a single month, while its share of spend stayed near 1%.
Anthropic’s share of spend grew from 61% to 65% in May, holding 70–80% of spend across every high-stakes use case (AI app generation, back office agents, and coding agents).
Cost-consciousness meant smarter routing between low-cost and frontier models. Customers got more deliberate about which model did which work, while overall usage kept climbing.
Last month, headlines about blown token budgets dominated tech news: Uber burned through its annual Claude Code budget shortly after Q1 and Amazon shut down KiroRank to curb unproductive tokenmaxxing. While runaway cost is a real problem, this month’s report shows that spend on production use cases still increased.
Two insights emerged from AI Gateway data in May:
Low-cost models entered production: New models shipped at price points that made the established labs look even more expensive, and they are capable enough to enter the mix in production.
Spend is increasing, but with smarter model mixes: Teams are still increasing token budgets, but they are implementing smarter routing strategies to get more value out of every dollar.
Low-cost models saw significant production volume for the first time
From February to April, volume distribution across labs on AI Gateway changed slowly, but in May, DeepSeek V4's launch completely shifted token share. The low-cost end of the market that barely existed in April became AI Gateway’s third-largest provider by volume in May, without a significant impact on overall spend.
In April, DeepSeek accounted for less than 1% of AI Gateway tokens and less than 0.2% of spend. In May, its volume share jumped to 17% of tokens, putting it in third place, ahead of OpenAI. Almost all of the volume comes from two models: deepseek/deepseek-v4-flash and deepseek/deepseek-v4-pro, both released in May.