Vercel Problems & Issues

Pro includes 1 TB of Fast Data Transfer per month, then charges from $0.15/GB (about $150 per extra TB), plus $2 per million CDN requests over the included amount. Cost write-ups show bandwidth and edge requests, not seats, driving most of the bill once a site passes ~250k monthly visitors, and bot or scraper traffic can trigger the same overages.

Pro now bills functions by Active CPU ($0.128/hour) and provisioned memory ($0.0106/GB-hour). Waiting on AI or database calls no longer counts as CPU, but instances kept warm by frequent cron jobs, bots or long streams still accrue memory charges. Community threads describe compute line items jumping several times over with no matching traffic increase.

Spend Management is now on by default for new Pro teams with a $200 on-demand budget and 50/75/100% alerts. But a budget alone does not stop usage: you must enable Pause Production Deployments, which takes every project offline, and Vercel checks spend only every few minutes, so charges can overshoot.

Pro is a $20/month platform fee that includes one deploying seat and $20 of usage credit; each extra developer (Owner or Member) seat costs $20/month. Since September 2025 read-only Viewer seats are free, which helps teams with reviewers, but a 10-developer team still pays $200/month before any usage.

Users report accounts being disabled within minutes of deploying projects, with automated emails citing fair use policy violations but providing no specific reason. Appeals go unanswered for weeks. Paid customers report losing access after spending $1,000+ with no recourse.

Hobby has hard caps (100 GB transfer, 1M function invocations, 4 CPU-hours, 5,000 image transformations, 100 deployments/day). Hobby cannot buy extra usage, so in most cases an exceeded limit means waiting 30 days or upgrading to Pro, and a viral post can take a free site offline.

Builds are cancelled when they exceed 45 minutes or the 8 GB memory limit of the default build machine (larger machines are Pro-only and billed per CPU-minute). The final log line is often just 'npm run build exited with 1', so the real error sits further up the log, and some failures (invalid vercel.json, unauthorized commit author) show no build log at all.

Next.js features such as image optimization, ISR cache revalidation and middleware work out of the box on Vercel but need extra adapters, cache handlers or configuration on other hosts. Teams that want to leave after a large bill often find migration takes real refactoring work.

On Hobby, deployments from a private GitHub repo are blocked when the commit author is not the Vercel account owner ('commit author does not have contributing access'). Vercel staff confirm this is expected: Hobby does not support collaboration on private repositories, so small teams have to upgrade or move the project.

Users report waiting over 30 hours for support responses, with some issues never resolved. Phone support is Enterprise-only. Support often responds with pre-canned responses without addressing specific technical problems. Can't talk to anyone when critical issues occur.

There is a Downgrade Plan button under Settings > Billing, but each account can only have one Hobby team, so downgrading while you already have one forces you to delete or merge a team. Downgrading also removes every member except the original owner, and connected stores and domains must be transferred manually first.

Vercel's free Hobby plan is for personal, non-commercial use only. Any project that earns money, including client work and ad-supported sites, must move to Pro, and commercial use on Hobby is a common reason accounts get paused. Many users only learn the rule after launch.

With Fluid compute, functions default to 300 seconds on every plan; Pro can raise maxDuration to 800s (1800s in beta), Hobby cannot. Slow database or third-party API calls, code paths that never return a response, and functions far from their data source still produce 504 timeouts, and the 1-hour Hobby log retention makes them hard to debug.

Scaling Next.js is problematic due to bad cache invalidation across replicas. ISR (Incremental Static Regeneration) cache invalidation works best on Vercel but causes issues on alternative deployments. Makes it easy to start but harder to grow.

In April 2026 Vercel disclosed that an attacker, entering through a compromised third-party AI tool, accessed internal systems and read non-sensitive environment variables for a limited subset of customers. Variables marked Sensitive showed no evidence of access, and Vercel advised rotating any secrets stored as plain variables.