The AI Cloud: A unified platform for AI workloads
By Steven Van ·
New AI Cloud primitives include AI SDK, AI Gateway, Active CPU pricing, MCP server support, Queues, and Sandbox for untrusted code.
Vercel has introduced the AI Cloud, a platform built on the same foundation as its Frontend Cloud but extended to support agentic workloads such as conversational AI frontends and autonomous agents, announced at Vercel Ship.
The AI Cloud brings together several AI-specific primitives:
- AI SDK and AI Gateway: AI SDK standardizes code across model providers so teams can swap models without changing code, while AI Gateway routes calls, manages API keys and provider accounts, and adds retries and fallbacks.
- Fluid compute with Active CPU pricing: a compute model that reuses existing resources before scaling and only charges compute rates while code is actively executing. Vercel says this can cut costs by up to 90% for workloads with high idle time, such as AI inference, agents, or MCP servers.
- Tool support, MCP servers, and Vercel Queues: agents can call tools, route calls to MCP servers via the Vercel MCP adapter, and offload or retry background tasks without blocking.
- Vercel Sandbox: runs untrusted, agent-generated code in ephemeral, isolated servers with user-accessible URLs, supporting runtimes including Node.js and Python with common packages pre-installed.
Vercel
The platform for frontend developers — deploy, preview, and scale web apps and AI agents with zero config.
View Vercel →