Renting is cheaper until you use the card about 20 hours a week for four years. An RTX 4090 costs about $1,600 plus $0.07 an hour to run, so it takes roughly 4,183 hours of real use to beat renting at $0.45 an hour. Most people manage four.
Everyone asks this question the same way. They look at the hourly rate, multiply it by a number they invent on the spot, compare it to the price of a card, and get an answer that is wrong because the invented number was wrong. The rate is not the hard part. Your own honesty about usage is the hard part.
This is a companion to our full guide on how renting a GPU actually works, and it exists to answer one question properly: at what point does buying the card become the cheaper decision?
The maths, shown rather than asserted
An RTX 4090 costs about $1,600. Running one pulls roughly 450 watts under load, which at the US average domestic electricity price of 15 cents a kilowatt-hour is about $0.07 an hour. Renting the same card is $0.45 an hour with the electricity included.
So the hourly gap between renting and owning is $0.45 minus $0.07. Every hour you use a card you own, you save that gap. Divide the purchase price by it and you get the number of hours it takes to earn the card back.
That is 4,183 hours.
Not per year. In total, ever, before the card is even level with renting.
See the numbers
| Use | Hours over 3 years | Renting | Buying |
|---|---|---|---|
| 2 hrs a week | 312 | $140 | $1,621 |
| 5 hrs a week | 780 | $351 | $1,653 |
| 10 hrs a week | 1,560 | $702 | $1,705 |
| 20 hrs a week | 3,120 | $1,404 | $1,811 |
| 40 hrs a week | 6,240 | $2,808 | $2,021 |
What that looks like in a calendar
Hours in the abstract mean nothing, so here is the same number as a habit.
| If you use it | You break even in |
|---|---|
| 40 hours a week | Two years |
| 27 hours a week | Three years |
| 20 hours a week | Four years |
| 10 hours a week | Eight years |
| 5 hours a week | Sixteen years |
The bottom two rows are the interesting ones, because that is where most people actually live. Sixteen years is not a payback period, it is a joke. The card will be a curiosity by then.
Put your own number in and see where you land:
How much would you actually use it?
At 6 hours a week, renting saves you about $1,242.
Buying assumes $1,600 for the card and $0.07 an hour in electricity. It does not count the computer you put it in, or the evening you spend installing drivers.
The number people get wrong
Ask someone how much they will use a GPU and they will answer with their best week. Not their average week, and certainly not the fortnight in August where they did not open the machine at all.
We see this constantly. Somebody sets out to generate product photos for a catalogue, works flat out for nine days, ships it, and then does not touch a GPU again for two months. Averaged across the quarter that is about three hours a week. During the nine days it felt like a full-time job, and it genuinely was, which is why the estimate comes out ten times too high.
If you want a realistic figure, look backwards rather than forwards. How many hours did you spend on this kind of work in the last three months? That number, divided by thirteen, is your honest weekly usage. It is usually a shock.
The three-year problem
There is a second clock running that the breakeven number ignores, and it works against buying.
A graphics card does not wear out, but it does fall behind. The card that comfortably runs everything today will, in three years, be the card that cannot fit the model everyone has moved to. This has happened repeatedly: 8GB cards were fine until image models grew, 24GB cards were fine until video work arrived, and each time a lot of people discovered their perfectly functional hardware had quietly become the wrong hardware.
That matters here because the four-year payback assumes the card is still the right card in year four. If it is not, you did not break even. You bought a card, used it for two useful years, and then went and rented something anyway.
Renting has the opposite property. When a new card lands in the fleet, you use it that afternoon at whatever it costs, and the old one stops being your problem.
What about buying used?
It genuinely narrows the gap, and it is the strongest version of the buying argument.
A previous-generation card at half price with most of its life left changes the arithmetic substantially. An RTX 3090 with 24GB — the same memory as a 4090 — can be found for a few hundred dollars second-hand, and at that price the breakeven drops to something like eight hours a week over three years. That is a real threshold that real people cross.
The catches are the ordinary ones. You are buying from a stranger, quite possibly a card that spent two years mining cryptocurrency at full load in a garage, with no warranty and no recourse. Some are fine. Some are not, and you will not know which until it fails in the middle of something.
If you are technical, patient, and willing to take that risk, buying used is a good decision and we are not going to pretend otherwise.
Where the maths stops mattering
Money is not the only thing being traded, and a piece that pretended otherwise would be a sales page.
A meter changes how you work. When the clock is running, people experiment less. They think twice before trying the sillier idea, which is sometimes the good one. A card you own has no meter, and for some people that freedom is worth more than the arithmetic says it should be. If you have noticed yourself doing this, it is a real argument for buying.
Owning means being the IT department. Drivers, thermals, a power supply that turns out to be 100 watts short, the model that needs a library version that fights with the one you have. This is a genuine cost and it is paid in evenings. Renting means someone else has already had that evening.
Waiting is a cost too. If the card you want is out of stock — and the good ones frequently are — renting is available this afternoon. Buying is available in six weeks.
The short version
Work out your honest hours per week over the last three months. If it is above 20 and you expect that to hold for years, buy the card. If it is below 10, renting is not close — it is the obvious answer and the gap widens the less you use it. Between 10 and 20, it is a genuine toss-up, and the deciding factor is usually whether you enjoy maintaining hardware.
Whichever way you land, start by finding out how long your actual work takes. The calculator on each model page turns a job into minutes and dollars, which is a better basis for this decision than any rule of thumb, including ours.