Vercel
by Vercel Inc.
The cloud that builds, deploys and runs AI applications and agents
Vercel is the managed cloud behind Next.js, now positioned as an 'AI cloud' for applications and agents. It gives engineering teams Git-connected deploys, preview environments, edge delivery and serverless compute, plus an AI layer — v0, the AI SDK, AI Gateway, Sandbox and durable agent runtimes — so product teams ship AI features without assembling their own inference and orchestration plumbing.
Vercel is the managed cloud built around Next.js, which it maintains, and since 2025 it has repositioned from 'frontend cloud' to 'AI cloud' — infrastructure for applications and agents that are themselves written and operated by AI. The core platform is unchanged in shape: Git-connected deployments, a preview URL per pull request, incremental static regeneration, a global edge network, a web application firewall with bot management, tenant isolation, and Fluid compute, which bills active CPU time rather than wall-clock invocation duration. Layered on top is an AI product line: v0, the natural-language app generator; the AI SDK, the TypeScript toolkit widely used for streaming chat and tool-calling interfaces; AI Gateway, a single endpoint and API key reaching hundreds of models across providers with automatic retry to a different provider on failure, spend observability, bring-your-own-key support and no markup on tokens; Vercel Sandbox for executing untrusted agent-generated code; Eve, a framework for durable agents; Passport, an identity layer for internal agents and deployments; and Containers for long-running workloads. Vercel was founded in 2015 as ZEIT by Guillermo Rauch, renamed in April 2020, is headquartered in San Francisco with roughly 550 employees, and closed a $300M Series F co-led by Accel and GIC in September 2025 at a $9.3B post-money valuation, following a $250M Series E at $3.25B in May 2024. It has acquired Turborepo, Splitbee, Tremor, NuxtLabs and Better Auth. Named customers include Shopify, Stripe, Notion, Ramp, Zapier and The Weather Company, the last serving 350 million monthly active users on the platform. Vercel holds SOC 2 Type 2, ISO 27001, PCI DSS 4.0 and TISAX AL2 attestations, with HIPAA support for enterprise accounts. The recurring criticism is not capability but cost predictability: usage is metered on six separate axes with no hard spend cap.
The VP Engineering or platform lead of a product team already standardised on Next.js or React who wants AI features and agent workloads shipped without building an inference gateway, a sandbox runtime and a CI/preview pipeline in-house.
A pull request becomes a live, isolated preview URL and a production deployment with no pipeline to maintain — and the same account provides model routing, sandboxed code execution and durable agent orchestration.
At a Glance
- Category
- Infrastructure & Cloud
- Pricing
- Freemium, Subscription, Usage-based, Contact for pricing
- Target Market
- CTOs, VPs of Engineering, Platform Engineers, Enterprise Developers, Product Engineering Teams
- Deployment
- Cloud-first, Multi-cloud
- Founded
- 2015
- Headquarters
- San Francisco, United States
- Team Size
- 500+
- Customers
- Not publicly disclosed; named customers include Shopify, Stripe, Notion, Ramp, Zapier, Avalara and The Weather Company
Key Features
- ✓AI Gateway
One API key and endpoint reaching hundreds of models across providers, with automatic retry to another provider on failure, spend monitoring, bring-your-own-key support and zero markup on tokens.
- ✓Fluid compute
Serverless functions billed on active CPU time rather than wall-clock duration, which materially cuts cost for I/O-bound AI workloads that spend most of their life waiting on a model.
- ✓Vercel Sandbox
Isolated ephemeral environments for running untrusted, agent-generated code, so a coding agent can execute and test its own output without touching production infrastructure.
- ✓Preview deployments
Every pull request gets its own immutable URL with the full application running, letting designers, PMs and stakeholders review a change before it merges.
- ✓AI SDK and v0
A TypeScript SDK for streaming chat, structured output and tool calling, plus v0, which generates working application code from a natural-language prompt.
- ✓Firewall, BotID and tenant isolation
Edge L3/L4 protection, DDoS mitigation, custom WAF rules, challenge mode and per-tenant isolation, with the OWASP Core Ruleset available on Enterprise plans.
- ✓Durable orchestration and Eve
A workflow runtime and agent framework for long-running, resumable processes including human-in-the-loop approval steps that must survive restarts and deploys.
Capabilities
Use Cases
- •Ship an AI chat feature without inference plumbing
A product team wires the AI SDK to AI Gateway, gets streaming responses and tool calling in days, and switches underlying models later without touching application code.
- •Run a coding agent's output safely
An internal developer platform executes model-generated code inside Vercel Sandbox, so a bad generation destroys a disposable environment rather than a shared build server.
- •Review changes before merge
Each pull request produces a live preview URL, so stakeholders test the real application and the team catches regressions before anything reaches production traffic.
- •Serve a global content site at scale
Editorial and marketing pages use ISR and edge caching so content updates propagate worldwide without a rebuild or a separately operated CDN configuration.
- •Orchestrate human-in-the-loop agent approvals
A durable workflow pauses an agent mid-task pending a human decision and resumes correctly hours later, surviving deploys and cold starts in between.
Ideal For
Best For
- ✓Teams building and deploying Next.js or React applications who want zero-config CI/CD and per-PR preview environments
- ✓Product teams adding LLM features who need one API key and automatic provider failover across hundreds of models
- ✓Agent builders who must execute untrusted, model-generated code in an isolated sandbox rather than on their own infrastructure
- ✓Marketing and content sites that need global edge delivery, ISR and image optimisation without operating a CDN
- ✓Engineering orgs that want durable, resumable workflow orchestration for long-running or human-in-the-loop agent approvals
Not Ideal For
- ✗High-bandwidth or traffic-spiky workloads on a fixed budget — usage-based metering has no hard spend cap, and Hacker News carries repeated reports of surprise invoices including a $96k bill on a viral app and $23k from a webhook denial-of-service
- ✗Teams that need portability guarantees — several Next.js features (image optimisation, ISR, middleware) are best-supported on Vercel, and moving elsewhere means adaptation layers
- ✗Organisations requiring true on-premises or air-gapped deployment; Secure Compute and BYOC are enterprise options but the control plane remains Vercel-operated
- ✗Cost-sensitive small teams paying per seat — Pro is $20 per team member per month before any usage overage
Integrations
Deployment
Market & Ratings
Not publicly disclosed; named customers include Shopify, Stripe, Notion, Ramp, Zapier, Avalara and The Weather Company
Market Analysis
Pros
- ✓Best-in-class developer experience for Next.js and React: Git push to global deploy with per-PR previews and no pipeline to maintain
- ✓The AI layer is unusually complete for a hosting platform — model gateway, sandbox, durable workflows and agent identity in one account
- ✓Strong compliance posture for a developer platform: SOC 2 Type 2, ISO 27001, PCI DSS 4.0, TISAX AL2 and HIPAA support on Enterprise
- ✓AI Gateway's zero-markup token pricing removes the usual reseller margin that model routers charge
Cons
- ✗Usage-based billing with no hard spend cap produces genuine bill shock — Hacker News threads document a $96k invoice on a viral app, $23k from a Stripe webhook denial-of-service, and a $3k bill from a single uncaught exception
- ✗Pricing has been re-modelled repeatedly since 2023, so a budget built on one year's meter dimensions does not survive the next
- ✗Meaningful lock-in around Next.js: image optimisation, ISR and middleware are best-supported on Vercel and need adaptation layers elsewhere
- ✗Per-seat Pro pricing at $20 per team member adds up before any usage — a ten-person team pays $2,400/year in base subscription alone
- ✗Suffered a security incident in April 2026 traced to a compromised third-party AI tool, exposing some employee credentials and environment variables
Pricing
Hobby
$0
- ✓Personal projects and prototypes
- ✓Capped usage with no option to buy more
- ✓Preview deployments
- ✓Global edge network
Pro
From $20/mo
- ✓$20 monthly resource credit then pay-as-you-go
- ✓Per-seat billing at $20 per team member
- ✓SAML SSO available as a $300/mo add-on
- ✓Advanced deployment protection as a $150/mo add-on
- ✓1 custom environment
Enterprise
Contact for pricing
- ✓OWASP Core Ruleset managed WAF
- ✓Audit logs and SCIM directory sync
- ✓SAML SSO included
- ✓Secure Compute and bring-your-own-cloud
- ✓12 custom environments
- ✓Platform SLA and dedicated support
Pro is $20 per team member per month and includes a $20 resource credit, after which usage is metered separately across edge requests, data transfer, compute (active CPU and provisioned memory), storage, operations such as ISR reads/writes and image optimisation, and build minutes. There is no hard spend cap, which is the source of the widely reported bill-shock incidents. SAML SSO ($300/mo) and advanced deployment protection ($150/mo) are paid add-ons on Pro and included on Enterprise; OWASP managed rules, audit logs, SCIM, Secure Compute and BYOC are Enterprise-only and unpriced publicly.
Security & Compliance
Connect
Sources
This page was written from 9 sources, 4 on domains other than vercel.com.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
ZML/LLMD
Free, Python-free LLM inference server that runs open models on NVIDIA, AMD, Google TPU, Intel and Apple chips from one binary
Lambda
GPU cloud and AI factories for training and inference — on-demand NVIDIA instances to single-tenant superclusters
Anyscale
Managed Ray platform for scaling AI data processing, training, inference and RL across thousands of GPUs on any cloud
Chroma
Open-source (Apache 2.0) vector and hybrid search database for AI, with a serverless Chroma Cloud on object storage