Choosing where a web application runs used to be a question of taste. In 2026 it is a question of fit, and of risk. Vercel changed how its functions execute, Heroku stopped developing new features, and AWS now starts new accounts on a credit-based free plan. If you are about to launch a customer portal, a SaaS product or an internal tool, the platform you pick now decides how you deploy, how you pay and how hard it will be to move later.
This comparison covers what each platform is built for, how their pricing models really work, what the Heroku announcement means for new projects, and the questions we ask before recommending one of them for a client build.
Heroku vs Vercel: what is the real difference?
Vercel started as a frontend cloud and remains strongest there: it builds your framework, serves pages from its global network and runs backend code as functions on demand. Heroku is a platform as a service that keeps your server processes, called dynos, running all the time.
That difference shapes everything else. On Vercel you push a Next.js, Astro, SvelteKit or similar project and the platform decides how each route is rendered and cached. Your API lives in functions that wake up per request. On Heroku you describe processes in a Procfile (a web process, a worker, a scheduler) and the platform keeps them alive behind its router, with add-ons such as Heroku Postgres attached through environment variables. AWS is neither: it is a catalogue of building blocks, from managed hosting to raw virtual machines, that you assemble yourself or with infrastructure-as-code.
- Vercel fits interfaces first: marketing sites, dashboards, storefronts and apps whose backend work is request-and-response.
- Heroku fits classic server apps: a Rails, Django or Express monolith with background jobs and a relational database.
- AWS fits anything, as long as someone on the team owns networking, permissions, monitoring and cost control.
What changed for Heroku in 2026?
On February 6, 2026, Heroku announced it is moving to a sustaining engineering model focused on stability, security, reliability and support, with no new enterprise contracts for new customers. Card-paying customers keep the same pricing and service, according to Heroku's own announcement.
In practice, nothing breaks for an app that already runs there. Dynos keep running, Postgres keeps backing up, and the dashboard works the same way. What changes is the outlook: a platform that is maintained rather than developed will not add the features competitors ship over the next few years. It follows an earlier change of course: the end of the free dynos and free data plans on November 28, 2022, explained in Heroku's "next chapter" post.
Our reading for a new project is simple. Heroku is still a reasonable home for an existing app, and it is still one of the simplest ways to run a worker next to a web process. For a product you expect to grow for years, keep the exit cheap: containerize the app, keep configuration in environment variables, and avoid add-ons you could not replace with a managed service elsewhere.
How does pricing compare across the three platforms?
The pricing models differ more than the headline numbers. Vercel charges per developer seat and then for usage, Heroku charges a fixed monthly price per dyno and per add-on, and AWS charges for every resource you consume, line by line. Prices below are as of October 2026.
- Vercel: the Hobby plan costs $0 and is restricted to personal, non-commercial use, while Pro starts at $20 per month per developer seat with included usage, as listed on Vercel's pricing page. A business site belongs on Pro from day one.
- Heroku: Eco dynos cost $5 per month (personal accounts only, and they sleep after 30 minutes without traffic), Basic dynos $7, Standard-1X $25 and Standard-2X $50, according to Heroku's pricing page. Databases and other add-ons are billed on top.
- AWS: a new account receives $100 in credits on sign-up and can earn up to $100 more, and the free plan account closes after six months or when the credits run out, as explained on the AWS Free Tier page.
What a production app actually costs depends on the parts that rarely appear in a comparison table: the database, bandwidth, log retention, preview environments, team seats and the hours someone spends operating the stack. For AWS, build the estimate in the official AWS Pricing Calculator with your real traffic, and add the engineering time. For Vercel and Heroku, multiply the per-seat or per-dyno price by what you will run in staging as well as in production, because preview and staging environments are where small teams tend to underestimate.
When is Vercel the right choice?
Vercel is the right choice when the product is mostly interface: a React or Next.js front end, content-driven pages and an API made of short requests. It removes server management and gives every branch its own preview URL.
Two details matter for backend work. First, functions have a time limit: on Hobby the maximum duration is 300 seconds, and on Pro the default is 300 seconds, configurable up to 800 (a 30-minute maximum is in beta), as shown in the Fluid compute documentation. For anything longer, the same documentation points to Vercel Workflows rather than one long function. Second, Fluid compute lets several requests share one function instance, which reduces cold starts for I/O-heavy work such as calling external APIs, and it has been enabled by default for new projects since April 23, 2025, according to the same page.
Where Vercel is a less settled fit: persistent WebSocket servers, long-running workers and scheduled batch jobs. WebSocket support in Functions, Daemons for long-running background processes and Queues were all still in beta in October 2026, and cron jobs invoke ordinary functions, with the same limits. If those parts are critical today, run them on a platform built for long-running processes, with Vercel serving the front end.
When does Heroku still make sense?
Heroku still makes sense for an existing app that runs well there, and for small teams that want a web process, a worker and a Postgres database without touching infrastructure. Its deploy model is simple and well documented.
The Procfile model remains one of the clearest ways to describe an application: one line per process type, scaled independently. Background jobs follow the pattern Heroku documents for worker dynos and queues, which keeps slow tasks away from web requests. If you start something new on Heroku in 2026, treat it as a deliberate, time-boxed choice and keep the application portable from the first commit.
When should you choose AWS instead?
Choose AWS when you need control the other two do not offer: private networking, specific compliance requirements, unusual workloads, or a scale at which paying for raw resources is cheaper than paying for convenience. Plan for the operations work it requires.
AWS gives you several levels of abstraction. Amplify Hosting covers front ends in a way that resembles Vercel. Elastic Beanstalk resembles Heroku for classic server apps. ECS on Fargate runs containers without managing servers, and Lambda runs event-driven functions. The flexibility is real, and so is the responsibility: identity and access policies, alarms, backups and budget alerts are yours to configure.
Can you combine Vercel, Heroku and AWS in one stack?
Yes, and many production apps do. A common split is the front end on Vercel, the API and workers on a container platform, and the database on a managed service. The cost is one more deploy pipeline and one more bill to watch.
A hybrid setup works when the boundary between the parts is a clean API. That is why the API style matters as much as the host: our comparison of REST and GraphQL for custom applications covers how that contract affects caching and performance. If you are still deciding how the pieces fit together, start with the guide to scalable application architecture, because the architecture should pick the platform, not the other way around.
How do we choose a platform for a client project?
We start from the workload, not from the brand. Five questions settle most cases: what runs for long periods, who will operate it, what data rules apply, what it costs at expected traffic, and how you would leave if you had to.
- Workload shape. Request-and-response pages and APIs point to Vercel. Long-running workers, sockets and schedulers point to a container platform or Heroku until Vercel's betas for them are generally available.
- Team. Without a dedicated DevOps person, managed platforms usually cost less than the hours AWS demands.
- Data and compliance. Healthcare, finance or contractual data rules often settle the question in favor of AWS networking and controls.
- Cost at your traffic. Model real usage, staging included, with each vendor's own calculator or pricing page.
- Exit plan. Containers, environment-based configuration and standard databases keep the next migration cheap.
When we build a custom web application for a client, the hosting decision is part of the architecture document, alongside the stack described in our React and Node.js production guide. And when the app has to talk to a CRM, a payment provider or an ERP, the platform has to support those connections reliably, which is the core of our API integration services.
The short version: deploy interfaces where interfaces are easiest, run long processes where long processes are cheap, and keep every layer portable. In 2026, that last rule matters more than any price list.