How much a mining farm makes in 2026: my $1,737 month, broken down by card

Over 31 days, from 3 August to 2 September 2026, my six machines with ten GPUs brought in $1,737 at an average utilization of 90%. This is not a back-of-the-envelope estimate or a screenshot of one lucky day: the income of every machine is published openly on the "Fleet" page of my site, Vast Sale, and refreshes every hour.

A word about the title first. I stopped mining a long time ago — the cards are rented out for other people's workloads on the Vast.ai marketplace. But the hardware is the same, the racks are the same, and the question I keep getting is the same: how much does a mining farm make today. So here are the numbers, what this sum is made of — and what is not in it.

August 2026: what each machine earned

The last column is not my asking price. It is one card's monthly income divided by all 744 hours of the month, idle time included. The rate a machine actually rents at is higher, because part of the time it sits empty. The rows add up to $1,736 — the one-dollar gap is rounding to whole numbers.

Since I joined the marketplace, the total has reached $23,638.

What the table shows at a glance

The average of $174 per card explains nothing here. The spread runs from $79 to $376 — almost fivefold. One machine with four 4090s made $906: that is 52% of the whole fleet's income from 40% of the cards. The three weakest machines together made $282 — less than a third of it.

Next come the two comparisons I put this table together for.

RTX 5090 vs RTX 4090: why the gap is not "several times"

$376 against $227 per card — 1.66×. Not twice, not three times.

A client pays for an hour of rental, not for raw performance. From there, everyone plugs in their own purchase price: if the new card cost noticeably more than the previous generation's flagship, it does not win that difference back in return per dollar invested. What it does give you is more money per slot in the chassis — and your slots and kilowatts are finite. That is the entire trade-off.

A separate note on the 4090 machine. It runs without virtual-machine mode: an RTX 4090 "dies" after being passed through to a VM — the reset bug, and no software fixes it. Clients who specifically need VM mode will never see this machine — and it still delivered half the fleet's income. Constraints like this are invisible in the table, but they are baked into it.

RTX 3080 Ti vs RTX 5080: the result I did not expect

The 3080 Ti made $114. The single 5080 made $79. A card two generations older earned 44% more than the new one.

I have no explanation I could prove with my own data, so I will not invent one. Income is the rate multiplied by paid hours, and the number of hours depends on price, bandwidth, the reliability score and the card class. Card class is only one of four variables, and not the main one.

The same story inside a single generation: the pair of 5080s in one machine made $86 per card, the single 5080 made $79. Here, though, I do know the reason, and it is not the hardware. On the paired machine I deliberately keep the price high: I am about to build a four-card 5080 machine out of these cards, and I do not want to hand the server over for a long rental right now — a client with a month-long contract would block the rebuild. At the same time I am watching whether there is any demand at that price at all. One of the two slots sits empty most of the month — and that is my decision, not the market's verdict on the card.

This is an important caveat for the whole table: a row's number is not the card's ceiling, but the result of my rate and my plans for that machine. If I wanted to squeeze out the maximum, I would set a market price and the row would look different.

The takeaway from this comparison is not "buy 3080 Tis". The takeaway is that the income gap between generations turned out to be far smaller than the price gap of the hardware, and you have to run the numbers on your own purchase.

Why I left mining for renting in the first place

I got into Ethereum mining back in 2017. In September 2022 Ethereum moved to PoS, and like everyone else I tried different coins — and fairly quickly concluded that this is a constant hopping from one project to the next. You find the one that pays more right now, reconfigure, then find another one and reconfigure again. The autumn of 2022 and the following year went into that.

I was already going to sell all the cards and forget the whole story. But exactly at that moment prices for used GPUs collapsed, and I did not want to give the hardware away for nothing. So I found myself at a fork: selling cheap hurts, holding for no reason is pointless. That left the third option — rent the card out instead of computing something on it.

The logic is simple. In mining you sell the result of computation — a coin whose price you do not control. In renting you sell the card's time. Same hardware, different business.

Clore: the first attempt and the first lesson

I started with Clore. The project looked unfinished to me, but the main problem was the payouts. At that time you could only get your earnings in the platform's own token or in bitcoin, and the fee was very high.

Think about what that means in practice. I was trying to escape volatility — and ran into it again through a different door: the money arrived in a token whose price I had no influence over whatsoever. Moving it into anything predictable cost a noticeable share of what I had earned. The conclusion I still live by: if the settlement currency jumps around, you do not have a stable income, no matter how nice the hourly rate looks.

One caveat: this is about how things worked when I joined. A lot has changed on the platform since, and if you are looking at it now — check the payout terms yourself, not through someone's two-year-old article.

Vast.ai: my review after two years

The choice fell on Vast.ai. The project is more serious and the client base is wider. By the summer of 2023 I was already looking in its direction, and almost a year went into studying the marketplace: how the price is formed, why one machine ranks above another in the listings, and what decides whether a client comes to you or walks past. My first machine went live on the marketplace in June 2024.

If you were looking for Vast.ai reviews from the host's side rather than the renter's — this is mine. There were difficulties; there is a whole section on them below. But after more than two years I can say it, and I will stand by it: for me this was the right decision. I have seen coins that paid more in the moment than my cards make in rental. Stability matters more to me.

How the money works on the marketplace

Three things worth understanding before the first machine.

The host receives 100% of their asking price. The client's markup is added on top, not deducted from your money. You get exactly the price you set.

Payouts are weekly, with a $20 threshold. Until the threshold is reached, the earnings stay on your balance.

Income = rate × paid hours. You set the rate yourself, and the hours depend on price, bandwidth, the machine's reliability and the card class. Only the first two are easy to control.

Mining farm payback: it is a method, not a number of months

I deliberately do not name a payback period — it is made of things I do not know about you: what you paid for the hardware, what a kilowatt-hour costs you, what your bandwidth costs.

The method is this: take a card's monthly income, subtract its electricity and its share of the bandwidth costs, and divide the purchase price by what is left. You get the period in months.

I will not name my own purchase price, and this is not secrecy. The fleet was assembled over several years, machine by machine, each at its own price, and hardware has gotten noticeably more expensive since. Any number I quote would be a lie either about today's market or about my real costs. Count from your own receipts — that is the only honest arithmetic.

What I can name precisely is the fixed costs. Per month I spend about 6,000 ₽ on electricity, 3,000 ₽ on the VPS for the tunnel, and another 750 ₽ on a backup VPS so that the main one going down does not take the whole fleet with it. About 10,000 ₽ in total. At an exchange rate of roughly 92 ₽ per dollar, August's $1,737 in revenue is about 160,000 ₽, so the fixed costs eat around six percent. The rest is hardware depreciation and my time, and everyone values those for themselves.

And the main thing: my $227 a month from one 4090 is August 2026 at my utilization and my rates. It is a reference for the order of magnitude, not a promise to you.

If you want to rent out a GPU: what to check beforehand

Practical points from my own experience — before you count any income.

• Bandwidth. It affects utilization no less than price does. In my case Russian ISPs were throttling the connection to the marketplace by the host's AS, and I had to set up a tunnel through my own VPS.
• Willingness to administer. Drivers, containers, networking. The list below shows what kind of tasks come up.
• A stable IP. Changing a machine's address penalizes the reliability score, and that score affects how often clients come to you.
• One machine to start. Watch it for a month: a single day shows nothing.
• Count after electricity. Counting before electricity is pointless.

Renting out a GPU is technically possible with almost any card. The economics does not start with just any card — see the $79 and $89 rows in the table.

What broke over these two years

Every failure is hours of downtime, that is, rows in the table that never got earned.

• The reset bug on the RTX 4090. The card does not survive being passed through to a virtual machine. The only fix is to give up VM mode.
• The VM support flag dropping after a marketplace software update — the reliability score collapses right after it.
• A stale CDI specification after a driver change: the machine vanishes from the listings with the status Not Listed. Everything looks fine from the outside, except nobody can see it.
• ISPs throttling by the host's AS. Hence the tunnel through my own VPS.
• Changing the IP address penalizes the reliability score.
• The marketplace controller getting stuck (heartbeat = 0). It is fixed by resetting its state, but first you have to notice it.

None of these problems solves itself.

Costs that are not in the table

$1,737 is revenue, not profit. The fixed costs I named above: about 10,000 ₽ a month for electricity and two VPS. But beyond them there is what does not fit into a monthly column: spare parts, wear on the cards, fans, disks — and my time, which goes into the list of failures from the previous section. I will not name a pretty "net" figure until I have tallied these costs as carefully as the income — and I do not want to pass revenue off as profit.

Who this is not for

• Those who want to set it up and forget it. The list above is not theory, it is my evenings.
• Those not willing to deal with drivers, containers and networking. Half the failures on that list are fixed by hand in a console.
• Those with one card in a gaming PC they need themselves. $79–114 a month from my weakest rows is unlikely to pay for the inconvenience.
• Those who need a fixed payment. Income depends on utilization, and utilization floats from month to month. 90% average utilization is a monthly average, not 90% every day.
• Those with poor or expensive bandwidth. On a weak connection the card will sit idle at any price.
• Those who need the maximum right now. Coins that pay more at this very moment appear regularly. I deliberately do not go there — if that trade-off annoys you, you will regret it.

If two or more items on this list apply to you — better not to start. I say this as someone who counts these numbers every month, not as someone trying to get you somewhere.

What next

One action: write to me and name your hardware, your electricity rate and your bandwidth — I will tell you which of my pitfalls apply to you and what to check before going live on the marketplace. My fleet's numbers for every machine are open on the "Fleet" page and refresh every hour, so there is something to check against. Write here: https://t.me/flux_vast