Funny timing - I was just on a call yesterday about renegotiating our enterprise Vercel contract. The Vercel employees on the call were very friendly, and did share information when prompted, BUT I came away from the call with the understanding that yep, their pricing is intentionally opaque. MIUs are 1 unit = $1, but the rate at which MIU are consumed vary by SKU. Which SKUs do you need, which are you using? Best of luck figuring that out. Cache hit? Fast Data Transfer. Cache miss? Fast Data Transfer _and_ Fast Origin Transfer, so 2x the cost.
For what its worth, they have an internal quoting tool, Copper, which we got a glimpse of on the call. This shows super detailed breakdowns of usage and pricing (for quoting, not actually for billing) and would be really useful to see...but of course they couldn't actually share that information with us.
Anyway, /rant. SaaS pricing being complex and not-exactly-user-friendly is nothing new.
This is a very naive take. It's so common to get big companies in as signal to others, or just match the current provider you have for X or offer you a cheap price to then hike it up next year, among many other sales tactics. It's definitely not "always ask for the most if the company has a lot". In fact I'd say companies with more money are more likely to get early good deals.
Have you done procurement yourself at the type of companies you describe?
No hate to any of the PaaSes out there, Vercel included. They truly serve a need.
That said, if you are an engineer planning on working in/around the field, I would strongly suggest developing some competence at basic Linux systems administration. (Also: learn SQL, even though it's out of fashion.)
Linux is probably the single technology where my knowledge has had the longest useful lifespan (SQL is probably second). There are Unix (System V) bits I learned decades ago that are still useful today, on Linux.
Then, you can use a PaaS if you want. But if it's not the right fit, you are in a position to do something else. You might find that designing your application with a modern compute stack (this is not a PaaS) gives you an unfair advantage.
There's now ~2 generations of professional engineers for whom SQL was rarely/never a thing to learn. Between the hard(er) split among front-end/back-end developers, ORMs improving, and the (flawed) idea that NoSQL would make SQL irrelevant, it has become somewhat of a niche skill.
Think about Firebase. One can be full stack on an app built on Firebase and be successful without ever touching SQL. Firebase is very popular, and has been for some time.
Source: I have worked with a set of otherwise solid engineering teams and can say that SQL familiarity has given me a leg up on very smart engineers who nonetheless do not do relational databases.
The only way you can get away with creating an application without touching sql is of you offload the logic to your backend language, and then I don't think you'd be efficient enough to scale.
Also can someone actually understand the logic of joins, indexes, pks, etc enough to create an efficientand scalable db, and not simply have learned sql by proximity?
> The only way you can get away with creating an application without touching sql
Please look at app platforms like Firebase[1]. There are absolutely complex Web applications running at scale that do not use SQL anywhere in the stack.
Aside from that, MongoDB and Redis are 17 years old; CouchDB is over 20. NoSQL is well-established at this point. All of the hyperscalers offer proprietary NoSQL databases, and have done so for years. A large number of developers uses those databases in production.
In our API-centric environment, there are a lot of apps that don't do much in the way of managing their own data directly at all, using mixtures APIs for auth and other key application functions.
> can someone actually understand the logic of joins, indexes, pks, etc enough to create an efficientand scalable db
If you are not using a relational database, these concepts do not necessarily apply.
I did some interviewing rounds with PaaS platforms for advisory roles. I loved Heroku and was ruined by Heroku so I thought maybe the industry has something new to offer.
The model is largely "built to be locked in" model. It is not something innovative. The issue is that to what scale of operation the platform considers there customer to be locked in where they can hand them a bill that compensates for the ease of entry model for all the hobby/free tier.
With Vercel I feel like these level is becoming lower and lower. You can within minutes launch a full startup with Vercel and AI assisted coding. And Vercel assumes that as long as you do not recieve any traffic that is good. The moment you recieve even a mild amount of traffic you are considered locked in.
To some people that is a fair trade because they have so little trust in their products in the first place investing hours instead minutes is a fair trade. If the traffic comes they are already in the green. If the traffic does not come any effort they have put in puts them in red. So, you put as little effort as possible to get thing out there.
I like Vercel. They have figured a monetization model for Slop SaaS. The other PaaS needs to catch up. In 2026, PaaS exists as a model to make revenue out of slop.
If you couldn't be bothered to write it yourself, why should I read it? The same goes for the overly-complex components that express the same idea over and over again, but somehow without adding any clarity.
The wheel you have to spin to have a chance of seeing a new paragraph is so uniquely aggravating it almost feels satirical, like those overcomplicated volume slider UI concepts people were making a while ago. [0]
I was looking at object storage recently and I hadn't realized how much profit cloud providers drive via egress. And it's so perfectly hidden from the marketing. Ended up going with Cloudflare R2 for free egress.
Just imagine if Huggingface used Vercel and paid their prices. It would cost them quite a few dollars for every decent-size model download. (Or AWS egress prices, or any comparable service.)
It is the business model to sponsor React, Next.js, the go to frontend and serverless deployment of enterprise headless SaaS products from Sitecore, Optimizely, AEM, SAP, Contentful, Sanity,... thanks to their partnerships that make Vercel the main option.
Vercel is similar to adopting Oracle, MS SQL Server, DB2,.... its use is decided at upper level, not what to use for weekend projects.
I'm not a Vercel fan, but the whole pitch of PaaS is you get more than just a provisioned server for your application. The 20$/month/dev is vercel's own concept of what that Dev Experience costs with a profit margin + average usage fees paid to AWS baked-in. They might leave that average low, but they assume you're here because you like vercel's tooling, not that you're price shopping for $/BitsTx'd. AWS will always win in that, because vercel is also AWS with some dipping mustards they really want you to lock into.
The hobby plan is a loss leader to get developers into the vercel tooling. If you go over the free tier's bandwidth limit, you've exceeded what vercel believes that developer goodwill is worth for a single account. If they allowed you to pay for extra bandwidth on your free plan, it would make vercel look like a crap cloud platform, because all you're doing at that point is paying a premium for AWS, and a kneecapped version of their developer tooling. They really want you to pay the 20$/month/dev and experience everything in vercel's platform because that's their only product. Honestly... no fault to them on that.
Maybe they'd gain some developer positivity about letting you dig your account out of the "exceeding the hobby limits" hole that's easy to fall into, but the AWS cost for them is already spent, and that was all the budget for appeasing you. You'll have to pay them to pay AWS anyway, so they draw a hard line at that point and demand you also pay to use the vercel tooling, which is the only thing they make. (or, in theory, telling you to go pay AWS yourself if the tooling is unimportant to you.)
Digitalocean's (not related to them in any way) app platform (and I'm sure many other cloud providers) provides almost everything that vercel does, at a fraction of the cost. I'm surprised this is not a well known fact.
This post appears to be AI generated (or heavily edited). eg “This wasn’t just about saving money.
It was about gaining control over how our system works”
Our first year on Vercel, the bill was $40,000. When our management went back to negotiate the second year, Vercel wanted $120,000! Vercel wasn't offering 3x the features, mind you, they just knew we were locked in. Our management got it down to $60,000 (still a 50% cost increase, year over year).
Our app is small beans, too. We don't even have that many users. To borrow a favorite term from DHH, Vercel are "merchants of complexity."
But they're only half the problem. Our management is the other half. They can't be bothered to grow a spine and move away from Vercel. So we'll just keep paying, and eventually some people will "be affected" by a "reduction in force."
off topic but I just wonder if this page is AI-designed. It looks quite good to my eyes. I feel like prior to coding agents this would instead be a blog post with some charts.
The only one that I would point out as better is Render, as it allows for containers, all others are worse than Vercel, in tooling, and supported languages for serverless on the backend.
I've had a great experience with cloudflare pages. It doesn't get much easier than using their cli (wrangler) to sync up a local folder. I suppose the exception is SSR, but then again I absolutely despise SSR so I don't think it counts.
I am saying this quite common and part of it is simply an enterprise tax. Yes vercel breaks it up into a separate charge but that’s their playbook and maybe they can better represent their costs and margin internally. I was only pointing out that SSO is almost always hated because enterprises will want/need it.
I've never been a Vercel customer and I know their offerings are not entirely comparable but that all seems way expensive compared to Cloudflare. Or Cloudflare plus any cloud service that can autoscale Docker containers and DBs.
I'm assuming the SSO charge is for access to Vercel admin and not end users.
I’ve been seeing this pattern a lot recently on HN. Single prompt preachy website designed to get attention.
Even if I’m in support of the subject matter (how many hours apple bugs waste for example), I think it’s generally in bad taste to be wasting the time of everyone on HN by fooling people with a website that’s meant to imitate something thought invoking.
My favorite Vercel pricing was the one where their AI token offering is just a wrapper around OpenRouter, and where OpenRouter has a few models for free, Vercel was charging for them.
Looking now, Arcee is no longer free, but the exact same tokens/model costs more on Vercel.
This will be the last month of my Vercel Pro Plan for now. I logged into Vercel just now to see what day of the month the billing period ends so I can move any projects or backup metadata before then, and when I clicked it, the page had the title "March 2026: Monthly Pro Plan". Needless to say, the invoice will not be for March, by any stretch. On the same page it says "This invoice will continue updating until the end of your billing period on May 20."
People aren't ranting about Vercel just because of aversions to trends or their marketing style. It's also because it has legitimately been buggy too often. A year ago I commented on HN about some other issues I experienced and that doesn't include weirdness with their open source or AI stuff. https://news.ycombinator.com/item?id=43588909
That said, I still think Vercel is a reasonable choice. Just not my top choice right now.
Serving 1M MAU for $1300/mo sounds like a reasonably good deal to me. I'm not a Vercel customer because I'd rather host my own infrastructure that isn't black-boxed by a bunch of abstraction I don't need. But no one should expect that $20/mo buys you much at all in the way of compute resources or bandwidth. You can't even get a home internet connection for $20 in a lot of places.
Not sure what the author is expecting -- a hosted site to be free forever no matter how big it gets? This site feels like someone grinding the dullest axe on the smallest possible wheel.
Funny timing - I was just on a call yesterday about renegotiating our enterprise Vercel contract. The Vercel employees on the call were very friendly, and did share information when prompted, BUT I came away from the call with the understanding that yep, their pricing is intentionally opaque. MIUs are 1 unit = $1, but the rate at which MIU are consumed vary by SKU. Which SKUs do you need, which are you using? Best of luck figuring that out. Cache hit? Fast Data Transfer. Cache miss? Fast Data Transfer _and_ Fast Origin Transfer, so 2x the cost.
For what its worth, they have an internal quoting tool, Copper, which we got a glimpse of on the call. This shows super detailed breakdowns of usage and pricing (for quoting, not actually for billing) and would be really useful to see...but of course they couldn't actually share that information with us.
Anyway, /rant. SaaS pricing being complex and not-exactly-user-friendly is nothing new.
Enterprise pricing always works out to “what can you pay? That’s exactly the price!”
This is a very naive take. It's so common to get big companies in as signal to others, or just match the current provider you have for X or offer you a cheap price to then hike it up next year, among many other sales tactics. It's definitely not "always ask for the most if the company has a lot". In fact I'd say companies with more money are more likely to get early good deals.
Have you done procurement yourself at the type of companies you describe?
No hate to any of the PaaSes out there, Vercel included. They truly serve a need.
That said, if you are an engineer planning on working in/around the field, I would strongly suggest developing some competence at basic Linux systems administration. (Also: learn SQL, even though it's out of fashion.)
Linux is probably the single technology where my knowledge has had the longest useful lifespan (SQL is probably second). There are Unix (System V) bits I learned decades ago that are still useful today, on Linux.
Then, you can use a PaaS if you want. But if it's not the right fit, you are in a position to do something else. You might find that designing your application with a modern compute stack (this is not a PaaS) gives you an unfair advantage.
> learn SQL, even though it's out of fashion
In what world is SQL out of fashion??
There's now ~2 generations of professional engineers for whom SQL was rarely/never a thing to learn. Between the hard(er) split among front-end/back-end developers, ORMs improving, and the (flawed) idea that NoSQL would make SQL irrelevant, it has become somewhat of a niche skill.
Think about Firebase. One can be full stack on an app built on Firebase and be successful without ever touching SQL. Firebase is very popular, and has been for some time.
Source: I have worked with a set of otherwise solid engineering teams and can say that SQL familiarity has given me a leg up on very smart engineers who nonetheless do not do relational databases.
The only way you can get away with creating an application without touching sql is of you offload the logic to your backend language, and then I don't think you'd be efficient enough to scale.
Also can someone actually understand the logic of joins, indexes, pks, etc enough to create an efficientand scalable db, and not simply have learned sql by proximity?
> The only way you can get away with creating an application without touching sql
Please look at app platforms like Firebase[1]. There are absolutely complex Web applications running at scale that do not use SQL anywhere in the stack.
Aside from that, MongoDB and Redis are 17 years old; CouchDB is over 20. NoSQL is well-established at this point. All of the hyperscalers offer proprietary NoSQL databases, and have done so for years. A large number of developers uses those databases in production.
In our API-centric environment, there are a lot of apps that don't do much in the way of managing their own data directly at all, using mixtures APIs for auth and other key application functions.
> can someone actually understand the logic of joins, indexes, pks, etc enough to create an efficientand scalable db
If you are not using a relational database, these concepts do not necessarily apply.
1 - https://firebase.google.com/docs/firestore/query-data/querie...
An easy way to host your own and fallback to a PaaS would eat the industry.
I did some interviewing rounds with PaaS platforms for advisory roles. I loved Heroku and was ruined by Heroku so I thought maybe the industry has something new to offer.
The model is largely "built to be locked in" model. It is not something innovative. The issue is that to what scale of operation the platform considers there customer to be locked in where they can hand them a bill that compensates for the ease of entry model for all the hobby/free tier.
With Vercel I feel like these level is becoming lower and lower. You can within minutes launch a full startup with Vercel and AI assisted coding. And Vercel assumes that as long as you do not recieve any traffic that is good. The moment you recieve even a mild amount of traffic you are considered locked in.
To some people that is a fair trade because they have so little trust in their products in the first place investing hours instead minutes is a fair trade. If the traffic comes they are already in the green. If the traffic does not come any effort they have put in puts them in red. So, you put as little effort as possible to get thing out there.
I like Vercel. They have figured a monetization model for Slop SaaS. The other PaaS needs to catch up. In 2026, PaaS exists as a model to make revenue out of slop.
If you couldn't be bothered to write it yourself, why should I read it? The same goes for the overly-complex components that express the same idea over and over again, but somehow without adding any clarity.
The wheel you have to spin to have a chance of seeing a new paragraph is so uniquely aggravating it almost feels satirical, like those overcomplicated volume slider UI concepts people were making a while ago. [0]
[0] https://news.ycombinator.com/item?id=27819384
AI;DR Vercel pricing is intentionally opaque and you are likely experience a massive bill, your site being taken offline, or both.
First time I’m seeing AI;DR. I think I’m going to be using that a lot.
I was looking at object storage recently and I hadn't realized how much profit cloud providers drive via egress. And it's so perfectly hidden from the marketing. Ended up going with Cloudflare R2 for free egress.
I go full course with supabase, they have compatible S3 storages. So auth, db, storage in same place and pretty easy to manage.
Just imagine if Huggingface used Vercel and paid their prices. It would cost them quite a few dollars for every decent-size model download. (Or AWS egress prices, or any comparable service.)
Vercel isn't Heroku.
It is the business model to sponsor React, Next.js, the go to frontend and serverless deployment of enterprise headless SaaS products from Sitecore, Optimizely, AEM, SAP, Contentful, Sanity,... thanks to their partnerships that make Vercel the main option.
Vercel is similar to adopting Oracle, MS SQL Server, DB2,.... its use is decided at upper level, not what to use for weekend projects.
I'm not a Vercel fan, but the whole pitch of PaaS is you get more than just a provisioned server for your application. The 20$/month/dev is vercel's own concept of what that Dev Experience costs with a profit margin + average usage fees paid to AWS baked-in. They might leave that average low, but they assume you're here because you like vercel's tooling, not that you're price shopping for $/BitsTx'd. AWS will always win in that, because vercel is also AWS with some dipping mustards they really want you to lock into.
The hobby plan is a loss leader to get developers into the vercel tooling. If you go over the free tier's bandwidth limit, you've exceeded what vercel believes that developer goodwill is worth for a single account. If they allowed you to pay for extra bandwidth on your free plan, it would make vercel look like a crap cloud platform, because all you're doing at that point is paying a premium for AWS, and a kneecapped version of their developer tooling. They really want you to pay the 20$/month/dev and experience everything in vercel's platform because that's their only product. Honestly... no fault to them on that.
Maybe they'd gain some developer positivity about letting you dig your account out of the "exceeding the hobby limits" hole that's easy to fall into, but the AWS cost for them is already spent, and that was all the budget for appeasing you. You'll have to pay them to pay AWS anyway, so they draw a hard line at that point and demand you also pay to use the vercel tooling, which is the only thing they make. (or, in theory, telling you to go pay AWS yourself if the tooling is unimportant to you.)
They will sell you pay-as-you-go services... but only once you pay their 20$: https://vercel.com/docs/limits#on-demand-resources-for-pro
Over all, I hate it. But I don't think there's anything too hidden about it, or at least no more than any other PaaS provider.
but... none of this justifies hiding price info...?
I mean, it would only justify being expensive
Just came across something relevant today: https://hanker.app/blog/how-hanker-cut-90k-a-year-by-moving-...
Digitalocean's (not related to them in any way) app platform (and I'm sure many other cloud providers) provides almost everything that vercel does, at a fraction of the cost. I'm surprised this is not a well known fact.
This post appears to be AI generated (or heavily edited). eg “This wasn’t just about saving money. It was about gaining control over how our system works”
Our first year on Vercel, the bill was $40,000. When our management went back to negotiate the second year, Vercel wanted $120,000! Vercel wasn't offering 3x the features, mind you, they just knew we were locked in. Our management got it down to $60,000 (still a 50% cost increase, year over year).
Our app is small beans, too. We don't even have that many users. To borrow a favorite term from DHH, Vercel are "merchants of complexity."
But they're only half the problem. Our management is the other half. They can't be bothered to grow a spine and move away from Vercel. So we'll just keep paying, and eventually some people will "be affected" by a "reduction in force."
off topic but I just wonder if this page is AI-designed. It looks quite good to my eyes. I feel like prior to coding agents this would instead be a blog post with some charts.
No doubt, these sites are now a dime a dozen. Flashy but really low signal to noise ratio, you can't unsee it.
Someone pls make something like this for Netlify
Alternative PAAS without the gotchas? Would appreciate proven alternatives. Thanks.
Coolify, Netlify, Cloudflare Pages (and Workers), Fly.io, Render, Railway...
Even GCP Firebase and AWS Amplify almost qualify as PaaS.
Cloudflare Workers are awesome. You can do a lot on the $5 tier. Their CLI tool `wrangler` is quite excellent.
Code generation tools know their APIs and they have excellent docs, so getting up and running isn't very difficult.
The only one that I would point out as better is Render, as it allows for containers, all others are worse than Vercel, in tooling, and supported languages for serverless on the backend.
None of which are nearly as beginner friendly as vercel.
And with Vercel you get to pay to stay a beginner forever.
I've had a great experience with cloudflare pages. It doesn't get much easier than using their cli (wrangler) to sync up a local folder. I suppose the exception is SSR, but then again I absolutely despise SSR so I don't think it counts.
Do they finally support native compiled languages like Vercel?
Webassembly doesn't count.
PSA: Railway just had this recent f*ck up - https://x.com/lifeof_jer/status/2048103471019434248
It's not Railway's fault that he didn't read the docs and thought it'd be a good idea to play Russian roulette.
lifeof_jer had this fuck-up, not Railway.
I'm going to plug Cloud 66 here for you as well. You get a happy middle between a fully managed PaaS and running your own servers.
It's fantastic that they let user purchase SSO as an add-on!
Not defending the practice but it’s really the enterprise tax.
no, with vercel it's an independent add-on. unique AFAIK.
I am saying this quite common and part of it is simply an enterprise tax. Yes vercel breaks it up into a separate charge but that’s their playbook and maybe they can better represent their costs and margin internally. I was only pointing out that SSO is almost always hated because enterprises will want/need it.
Edit not sure why I typed hated. I mean always upcharged.
for what it's worth the creator mentioned this site was made in a couple of minutes lol
I've never been a Vercel customer and I know their offerings are not entirely comparable but that all seems way expensive compared to Cloudflare. Or Cloudflare plus any cloud service that can autoscale Docker containers and DBs.
I'm assuming the SSO charge is for access to Vercel admin and not end users.
Pure slop page, even by recent HN submission standards. Straight from Opus' mouth, minimal editing done.
I’ve been seeing this pattern a lot recently on HN. Single prompt preachy website designed to get attention.
Even if I’m in support of the subject matter (how many hours apple bugs waste for example), I think it’s generally in bad taste to be wasting the time of everyone on HN by fooling people with a website that’s meant to imitate something thought invoking.
Excellent work - thank you!
Claude worked hard on it for many minutes!
Tokens aren’t free, you know!
My favorite Vercel pricing was the one where their AI token offering is just a wrapper around OpenRouter, and where OpenRouter has a few models for free, Vercel was charging for them.
Looking now, Arcee is no longer free, but the exact same tokens/model costs more on Vercel.
https://vercel.com/ai-gateway/models?providers=arcee-ai
https://openrouter.ai/provider/arcee-ai
arcee-ai/trinity-large-thinking
Vercel: $0.25/M $0.90/M
Arcee: $0.25/M $0.80/M
This will be the last month of my Vercel Pro Plan for now. I logged into Vercel just now to see what day of the month the billing period ends so I can move any projects or backup metadata before then, and when I clicked it, the page had the title "March 2026: Monthly Pro Plan". Needless to say, the invoice will not be for March, by any stretch. On the same page it says "This invoice will continue updating until the end of your billing period on May 20."
People aren't ranting about Vercel just because of aversions to trends or their marketing style. It's also because it has legitimately been buggy too often. A year ago I commented on HN about some other issues I experienced and that doesn't include weirdness with their open source or AI stuff. https://news.ycombinator.com/item?id=43588909
That said, I still think Vercel is a reasonable choice. Just not my top choice right now.
Edit: by the way, remember now.sh? https://x.com/vercel/status/717764348706316288 It's funny that ten years later, there is a similar service with a similar name, exe.dev
Am I getting old or can you guys actually read this site? The text is tiny and gray...
No, I feel the same way.
Serving 1M MAU for $1300/mo sounds like a reasonably good deal to me. I'm not a Vercel customer because I'd rather host my own infrastructure that isn't black-boxed by a bunch of abstraction I don't need. But no one should expect that $20/mo buys you much at all in the way of compute resources or bandwidth. You can't even get a home internet connection for $20 in a lot of places.
Not sure what the author is expecting -- a hosted site to be free forever no matter how big it gets? This site feels like someone grinding the dullest axe on the smallest possible wheel.
The author isn't expecting anything. Opus might be, as that's what wrote eveyrhting.