A cloud GPU list where the free-credit column means six different things and the prices are typed by hand
List of Deep Learning Cloud Providers
At a glance
- What is it?
- Deep-learning-in-cloud is three files and one long markdown table of GPU vendors, deployment hosts and MLOps platforms. Useful as an index, but every price and every free offer in it is hand-written text with no date attached and nothing that could check it.
- Who is it for?
- Use this list for what it is good at, which is telling you the names of the vendors nobody has heard of, and verify anything you are about to budget against the vendor's own page. The free credit amounts are the most trustworthy part, because they are specific and come from the large providers: 300 dollars from three of them, 200 from two, a free Kubernetes control plane from one, and a 166 hour GPU grant from another.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- GitHub does not report a main language for this repository.
Answers come from the project's GitHub data, last synced on October 9, 2026, and from our analysis. They are not legal advice.
Editorial analysis
Three files, and no mechanism that could age a price
The repository is `.github/`, `LICENSE` and `README.md`. There is no data file, no script, no site generator and no workflow that fetches prices.
That matters more than it sounds. A vendor list of this kind looks like data and behaves like prose, so the only way a price changes is for someone to open the markdown and type over it. There is no timestamp per row, no last-verified column and no history that maps a price to a date other than the commit that changed the line.
The consequence is that a row is not old or new, it is simply whatever the text says now. A per-hour figure for a specific card is the kind of number that moves often, and a credit promotion is the kind that expires quietly without the page changing.
The tables are also hand-formatted rather than generated, which shows up in the cells themselves. Several price cells contain a placeholder that is a GitHub emoji shortcode rather than text, and one free-credit cell ends with a stray fragment of punctuation after its dash, the residue of an edit that was never cleaned up.
One document doing three jobs, with tables that differ in size by five times
The list covers cloud GPUs for training, hosts for serving a model as a web app, and MLOps platforms. That is a reasonable scope for someone starting out, and it is the reason the tables drift out of sync with each other: a serving host has no GPU price, and a GPU provider rarely has a free plan.
The sizes are very different. The training table runs to roughly thirty-four vendors. The deployment table has seven. The third section, on MLOps platforms, introduces itself with a sentence that stops mid-clause, and no rows for it appear in the document where the two tables above are written.
So the section aimed at teams already past experimentation, described as being for serious enterprise level work, has the least to read of the three. That ordering is defensible for a personal list whose author started where most people start. It is less defensible as a reference, because a reader's need and the table's length are inversely related.
The other thing a shared document does badly is mix units of comparison. One table is sorted alphabetically and treats a price per hour as its sort of meaning; the other has no shared metric at all. Neither is wrong, but a reader has to work out for themselves which column answers their question.
The free trial column holds credits, hours, plans, betas and referrals
The fourth column is headed free trial or free credits, and what it contains varies by row.
Credit amounts are the cleanest entries. Three vendors are listed at 300 dollars, three more are not, with two at 200 dollars, one small provider at a free 10 dollars worth, and a 100 dollar credit that is conditional on the GitHub student pack rather than open to anyone. Genesis Cloud is given as 166 free GPU hours, which is the only row expressed in hours.
Then the column changes meaning. One entry is a free tier defined by an accelerator stack rather than by time or money: ROCm notebooks on AMD hardware, described as Colab-style. Another is not compute at all but a free Kubernetes API with a limit of ten GPUs per pod, which is a free control plane attached to paid machines. A third is a free plan for a serving host, and a fourth is a product marked as currently in beta.
One row advertises a referral program, misspelled in the cell, which is the only place in either table where the list is compensated rather than descriptive. And at least six rows put a dash in both the price and the free column, meaning no information at all.
A third of the rows carry a price, and the units do not line up
Where a real number appears, it is usually in place of a placeholder string that reads `:label:`, which is a GitHub emoji shortcode that never renders as anything useful. That placeholder sits in most of the GPU rows.
The rows that do have numbers use at least four different units. Most are per hour in dollars. One is billed per second in Indian rupees. One is per month in euros, for dedicated GPU servers rather than an hourly instance. Another sells a fraction of an A100 per hour in euros rather than a whole card.
Even within the hourly dollar rows the comparison does not hold. The same A100 appears at 0.55 dollars an hour at one vendor, 1.10 at another, and around 2.35 for the 80GB part at a third. Within a single vendor, a 4090 at 0.34 and an H100 at 1.99 are both listed, which is a sixfold spread that reflects the card rather than the vendor.
The practical result is that the price column is good for finding out which vendors publish numbers at all, and useless for choosing between them. Three of the visible figures also come with qualifiers rather than plain amounts: one rate is marked as starting at, and one entry notes a per-second billing model alongside its hourly equivalents.
Two free-forever claims carry an asterisk that is never resolved
The first two rows are notebooks rather than rented GPUs, and both are marked free forever with a footnote marker attached to the word forever.
No annotation for that marker appears in the text between the tables and the sections that follow. The introduction does point readers to a perks and offers section for finding free GPU hours, so the explanation may be intended to live there, but the marker itself is unresolved where the claim is made.
That is worth pausing on, because the two claims are the ones most likely to be quoted. A free tier that is free until it is not is a different proposition from a free tier with a stated daily limit, and the difference is exactly what a footnote would settle.
The same column also holds the reverse case, where a vendor is listed with a price and no free offer at all, which is a dash rather than the word none. Six or more rows are in that state. So a reader scanning the column sees a mix of claimed permanent free access, unstated terms, and silence, with no column that says which is which.
A vendor table that includes things that are not vendors, and one link that tracks itself
The first table is headed cloud vendor, and its rows range from hourly GPU hosts to adjacent products. One entry is a product whose status in the price column is currently in beta. Another is an enterprise machine learning platform rather than a place to rent a card by the hour. A deployment platform appears in the same table as the training providers.
The result is a table where the pricing column does not mean one thing, which is a different problem from the free column. A dash in the price column can mean no hourly GPU product, a price page behind a label, or a vendor the person adding the row never checked.
One row is worth noting for a different reason. Its link carries tracking parameters naming the source as GitHub, the medium as this repository, and the campaign as a list of deep learning cloud resources. That is the author tagging their own link, which is harmless and transparent, and it is the only row where the provenance of a referral is visible in the URL itself rather than only in the free-credit cell.
The other row with a commercial angle is the one advertising a referral program, and it sits in the same column where a 300 dollar credit would otherwise appear, so the two are not visually separated.
The only technical limit in either table is attached to a free plan
The only hard constraint in the document appears in the deployment table, and it is a size limit.
A serving host's free plan is described with a model size limit of under 500 megabytes, and nothing else in either table states a number that constrains what you can run. The neighbouring free column entries are a free beginner account, a free plan, a student-pack credit and a beta marker, none of which tell you whether your artefact fits.
That asymmetry is the practical gap. Model serving fails on artefact size far more often than it fails on price, and the list has an answer for the price question in one row and for the size question in one row, from different vendors.
The deployment table's own shape makes the point too. It is seven vendors wide, and two of them carry no price at all, one is marked as currently in beta, and the remainder differ on every axis the reader cares about: credit size, student eligibility, and monthly cap.
Editorial conclusion
Use this list for what it is good at, which is telling you the names of the vendors nobody has heard of, and verify anything you are about to budget against the vendor's own page. The free credit amounts are the most trustworthy part, because they are specific and come from the large providers: 300 dollars from three of them, 200 from two, a free Kubernetes control plane from one, and a 166 hour GPU grant from another. The per-hour prices are the least trustworthy, because they use at least four different units and only a minority of rows have one at all. Before you commit, check three things. The free tier's shape, since the column mixes credits, hours, plans, beta products and referral programs, and only one row anywhere states a technical limit, a half gigabyte model cap on one free plan. Whether the vendor you pick is a GPU host at all, since the same table holds a product marked as currently in beta and rows with no data in either column. And the age of the number, which is the date someone last edited that line, since nothing in the repository can tell you.
Frequently asked questions
What is the zszazi/Deep-learning-in-cloud repository?
A hand-maintained markdown list of cloud GPU vendors, model deployment hosts and MLOps platforms, with a pricing column and a free trial or credits column for each. It is three files: a licence, the README, and a GitHub directory. There is no data file, script or workflow behind the tables.
Which free GPU credits does the Deep-learning-in-cloud list name?
Three vendors are listed at 300 dollars, two at 200 dollars, one small provider at a free 10 dollars worth, and one at 100 dollars conditional on the GitHub student pack. One row is given as 166 free GPU hours, one as a free Kubernetes API with up to ten GPUs per pod, and one advertises a referral program rather than credits.
Does the Deep-learning-in-cloud list say how much a GPU costs?
For roughly a third of the GPU rows, and the units differ: per hour in dollars for most, per second in Indian rupees at one, per month in euros at one, and a fraction of an A100 per hour in euros at another. The remaining rows show a placeholder string instead of a number, and no row carries a date.
Can I serve a model for free using a host from this list?
The deployment table names a free plan at Heroku with a model size limit under 500 megabytes, a free beginner account at PythonAnywhere, and a free plan at Vercel. Glitch and Render are listed with no price given, and Streamlit for Teams is marked as currently in beta.
How do I know if a price in the Deep-learning-in-cloud list is current?
There is no way to tell from the file. The tables are prose in a single README, nothing fetches or checks the prices, and no row is dated. The only way to date a figure is the commit that last edited that line, so a price should be confirmed on the vendor's own page before it goes into a budget.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/zszazi-deep-learning-in-cloud)