Vercel CEO Guillermo Rauch: 1 trillion tokens a day and the split between AI models and agents

On July 6, 2026, in an interview with TechCrunch, Vercel CEO Guillaume Rauch revealed the scale of the platform's growth and explained why companies that have moved into production with AI applications are inevitably transitioning to a multi-model architecture with a separation of model layers and agents.

What Vercel's Numbers Show

Vercel processes 6 million deployments daily — and half of them are initiated by agents rather than humans. Through the platform's AI gateway, more than 1 trillion tokens pass in a single day.

These figures reflect not a forecast, but an already accomplished shift: agents have stopped being an experimental tool and have become part of the core infrastructure flow. According to Rauch, two "killer applications" for agents have emerged today:

• Coding agents — the main driver of token consumption on the platform
• Enterprise agents — systems for automating internal processes while accounting for security and compliance requirements

Why Production Changes Architectural Decisions

Rauch directly names the main trigger for architectural changes.

"When you optimize for production, you start looking at price/performance."

While an AI product remains a prototype, a developer can work with a single provider for convenience. In a production environment, the logic changes: different tasks within an agent scenario require different cost-to-power ratios. One agent step may require deep reasoning, another a fast and cheap call. Rigid attachment to a single model makes optimization impossible.

This is why Vercel's client companies are now combining providers — OpenAI, Anthropic, Google Gemini, and open-source models like DeepSeek — depending on the characteristics of the task, not out of loyalty to a single lab.

This only works with clear architectural separation: the agent layer (orchestration, tools, memory, prompt logic) exists independently from the model layer. Swapping out the model under the hood doesn't affect agent logic.

New Vercel Tools for Production Agents

By the date of the interview, Vercel released two new products:

• Eve — a framework in which agent instructions and capabilities are described in natural language instead of code
• Vercel Sandbox — an isolated environment that restricts agent access to data while preserving their computational capabilities

Both products are aimed at making agent infrastructure manageable and predictable in production. Rauch also speaks openly about competitive pressure: major AI labs themselves are moving into adjacent niches — hosting, deployment, infrastructure. Vercel's answer is a bet on open protocols and positioning as an independent infrastructure layer that doesn't depend on any one lab.

What This Means

With more than a trillion tokens per day, Vercel is seeing the market transformation earlier than most analysts: the shift from a single-provider strategy to multi-model architecture is already an observable pattern. Rauch positions Vercel as the "AWS of this generation" in AI infrastructure — a platform on which applications are built on top of any models.

Frequently Asked Questions

How many tokens does Vercel process per day?

More than 1 trillion tokens pass through Vercel's AI gateway daily. Moreover, half of the 6 million daily deployments are initiated by agents rather than humans.

What is Eve from Vercel?

Eve is a new Vercel framework that allows you to describe agent instructions and capabilities in natural language instead of code. The tool was introduced in July 2026.

Why did Vercel release Sandbox?

Vercel Sandbox restricts agent access to data while preserving their computational capabilities — this is a response to enterprise clients' demands for security and manageability in a production environment.