Why Self-Host Next.js Apps?

There are many ways to run your Next.js apps, and the most important and primary one is, of course, from the creators of Next.js, hosting it at Vercel. Running your prototype or newly created app with Vercel is very straightforward, but when you run it in production, costs and complexity quickly escalate.

For starters, the Hobby plan does not allow you to run commercial apps, and the Pro plan starts from $20/month per user. If you exceed execution or bandwidth limits, it can easily cost $60+ per month.

Serverless

Vercel advocates and implements a so-called serverless way of running Next.js apps. In practice, it means that frontend code after the build is cached globally, and backend is packaged to quickly scale as needed and run where it is close to the users (so-called edge computing).

While certainly this brings speed improvements to the properly written apps, in practice, most of the businesses, except truly global ones, do not need edge presence and can decide on one specific region where they can host their apps

But this comes at a significant cost & complexity of the app - it is challenging to run specific workloads on this infrastructure, the execution times and available memory are limited, forcing you to rely on external services or even additional backend for more feature support.

Other serverless providers

Thanks to the OpenNext initiative, other providers can also deploy your app as serverless:

• Netlify
• Cloudflare

They involve similar trade-offs to Vercel: execution is bounded by the platform's limits, and configuration for non-trivial apps can get complex.

Usage-based serverless billing also carries the risk of a large, unplanned bill from a traffic surge, a misconfiguration, or rogue AI bots. Documented cases run into tens of thousands of dollars (see our breakdown of a $46k Vercel bill).

Comparing serverless to self-hosting

Compared to serverless and self-hosting, it is a bit more complicated at the start, and has a flat pricing of renting the server from a cloud provider. In return, it gives you a fully functional platform to scale your app and run different workloads.

Why self-hosting?

Here are some benefits to self-hosting Next.js, compared to using serverless. They are more important as your app becomes more complex, with more features and backend functionality.

Vercel free (hobby) plan limitations

• By ToS, you can't use hobby plan for commercial purposes. Vercel might detect if you have Google Ads or Google Analytics etc running on it. It seems to be rarely enforced now.
• You cannot use private repositories in the GitHub organisations (only private personal repos)
• Limited traffic and backend operations

Hard-capped pricing

Yes, if you get hit by the traffic or a wrongly written AI bot, your server will stop working, without costing you a thousand-dollar bill. And many traditional PaaS providers like Hetzner and Digital Ocean have much more generous traffic limits, in terabytes, not gigabytes.

Artificial Intelligence Fuels Rise of Hard-to-Detect Bots That Now Make up More Than Half of Global Internet Traffic, According to the 2025 Imperva Bad Bot Report | Business Wire

Thales, the leading global technology and security provider, today announced the release of the 2025 Imperva Bad Bot Report, a global analysis of automated bot traffic across the internet. This year’s report, the 12th annual research study, reveals that generative artificial intelligence (AI) is revolutionizing the development of bots, allowing less sophisticated actors to launch a higher volume of bot attacks with increased frequency. Today’s attackers are also leveraging AI to scrutinize their unsuccessful attempts and refine techniques to evade security measures with heightened efficiency, amidst a growing Bots-As-A-Service (BaaS) ecosystem of commercialized bot services. This press release features multimedia. View the full release here: https://www.businesswire.com/news/home/20250415432215/en/ ©Thales Automated bot traffic surpassed human-generated traffic for the first time in a decade, constituting 51% of all web traffic in 2024. This shift is largely attributed to the rise of A

Home

Bigger traffic limits

Traffic limits comparison. List prices and platform limits checked September 2026; verify against each provider's pricing page and vercel.com/docs/limits.

No execution time or memory limits

When you run your app on the server, there are no inherent limits set by the framework. You can limit your execution yourself, or memory is limited only by the amount of RAM on your host, or how much you allocated memory to your app.

Also, response limit payload from the backend is not limited, as opposed to Vercel, where it is max 4.5Mb

How to allocate more memory to your app

JavaScript applications can run into memory-related issues as they scale, and understanding how to manage memory effectively is crucial for building robust applications. One of the most important concepts in Node.js memory management is the max-old-space-size parameter, which directly impacts how your application handles memory allocation. What is max-old-space-size?

DollarDeployRuslan Gainutdinov

Scheduled tasks support

You don't need to use an external service for task scheduling - use the cron format and ping your backend endpoint to execute the cron job.

These cron jobs can be as long running as you want, as opposed to max 60s on Hobby plan and 300s on Pro plan.

Websocket and SSE support

Websocket or SSE (server-side events) is a popular way to add interactivity to your app, for example, for AI chat, or other.

When you self-host, you don't need to do anything special (besides configuring Nginx or your reverse proxy properly). On Vercel's serverless functions you need special configuration or a third-party; Cloudflare supports WebSockets through Durable Objects rather than plain serverless functions.

Queues and background jobs

Long-running background workers are constrained on serverless. Vercel offers Cron Jobs and queue/background functions, but always-on workers and long jobs fit a persistent server better.

When self-hosting, you can implement that using BullMQ Node.js library and Redis for powerful, full-featured background jobs processing.

PDF generation

Because of memory and time constraints, it can be difficult to generate a PDF on the backend. One common way to do it is to run Puppeteer on the server, but this is impractical on serverless functions, where headless Chrome is constrained by deployment size, memory and execution-time limits.

If you self-host, you can just run this normally, as regular code, just make sure you have all dependencies installed.

Screenshot or website scraping

This is impractical on serverless for the same reason as Puppeteer above: it needs Puppeteer or Playwright driving a real Chrome browser, which the serverless size, memory and time limits make hard to run.

Using a third-party is always an option, but at that moment, your app looks like a patchwork of services, not an independent project.

Running your LLMs

Large language models and your own AI need a lot of memory to run, sometimes also GPU support on the host. This is not feasible on Vercel or serverless functions; you either call hosted APIs (OpenAI/Anthropic/Perplexity) or run the model on a server with enough RAM or a GPU.

Conclusion

Using the right tools for self-hosting makes your costs predictable and architecture simpler, as you don't need to rely much on third-party services.

Try using DollarDeploy - it makes deploying Next.js apps easy. And you can choose which hosting provider to use, and with our Digital Ocean and Hetzner integration, it is very easy to create a new server and run your Next.js app on it.

Try DollarDeploy now!