Rauch is describing a structural shift already underway: companies stopped picking one AI lab and started treating models as swappable components. For SaaS founders building on AI infrastructure, this is the strategic question worth sitting with.
- Vercel now sees 6 million deployments a day, half triggered by coding agents, with over 1 trillion tokens flowing through its AI Gateway daily.
- CEO Guillermo Rauch told TechCrunch companies have moved from picking one AI lab to treating models, agents, and data platforms as separate, swappable pieces.
- Rauch pointed to Gemini’s price-performance as a driver of adoption, alongside rising use of open models like DeepSeek and GLM-5.2.
- Vercel built two internal tools, Eve for agent instructions and Vercel Sandbox for data access control, after learning agents in production create new security and data leakage risks.
- Rauch sees direct competition forming with AI labs as they add infrastructure features, calling it a fight over whether models and agents stay coupled.
What Happened
Vercel CEO Guillermo Rauch told TechCrunch that companies have shifted from picking a single AI lab to treating models, agents, and data platforms as separate, swappable components. The company now sees 6 million deployments a day, half triggered by coding agents.
More than 1 trillion tokens flow through Vercel’s AI Gateway daily. Rauch said clients are increasingly optimizing for price-performance rather than brand loyalty to one lab, driving growth in Gemini alongside open models like GLM-5.2 and DeepSeek.
Vercel built two internal tools in response to production risks it hit at scale. Eve lets teams define agent instructions and skills in natural language, while Vercel Sandbox restricts what data an agent can access or send externally.
Rauch cited a conversation with Airbus leadership about coding tools accidentally training on proprietary aerospace code as the kind of risk driving these controls. He also described an internal sales agent built on Eve that answers account-growth questions instantly instead of waiting on quarterly dashboard projects.
Why It Matters
Rauch’s framing cuts to a real infrastructure decision every SaaS team building on AI now faces: whether to commit to one lab’s full stack or build on swappable components across labs. His bet is that decoupling wins, positioning Vercel as connective infrastructure rather than a single-vendor dependency.
Rauch has an obvious incentive to describe the market this way, Vercel’s entire business model depends on models and agents staying decoupled rather than converging inside one lab’s platform. His own example, OpenAI shipping website publishing tools that compete directly with Vercel’s hosting business, shows the labs are already moving toward the opposite outcome he’s betting against.
“The reality is, when you’re optimizing for production, you start looking at a price-performance.”
Guillermo Rauch, CEO, Vercel
Bottom Line
Watch whether major AI labs keep expanding into infrastructure territory, website hosting, deployment, data platforms, at the pace OpenAI just did. Each move in that direction weakens the decoupled future Rauch is describing.
For SaaS founders, Rauch’s multi-model production data is the more actionable takeaway than his infrastructure thesis. If your team is still locked into a single AI lab by default, price-performance testing across models is now standard practice, not a risky move.
