GPU Rent HubMonthly vs hourly

Monthly vs hourly GPU rental: where the crossover actually is.

GPU Rent Hub rents dedicated NVIDIA H100 SXM servers by the month for $1,690 flat, which is $2.35 per GPU-hour across a 720-hour month, paid in Bitcoin, Ethereum, USDT, USDC, Monero, Solana or Litecoin with no identity verification. Typical US on-demand hourly pricing for the same card in 2026 runs $2.49 to $3.49. Here is the arithmetic, including the cases where you should use an hourly provider.

$2.35H100 SXM, per GPU-hour on a month

720 hHours in a monthly term

55–70%Utilisation where monthly wins

20 TBOutbound included, per instance

01 Definition

What is the difference between monthly and hourly GPU rental?

Two products that both hand you a GPU. The billing model changes what you optimise for.

Monthly GPU rental is a fixed-price term lease of a dedicated GPU server. You pay one flat amount for 30 days of exclusive access to specific physical hardware, and the price is the same whether the card runs flat out or sits idle. Hourly GPU rental is metered access to a shared pool. You pay per hour or per second of runtime, usually plus separate line items for persistent storage and outbound data, and you release the card when you stop. Monthly costs less per hour of real work above roughly 55–70% utilisation. Hourly costs less below it.

GPU Rent Hub sells weeks and months. The shortest term is seven days. There is no per-second meter and no plan to build one.

02 The table

How much does an H100 cost per hour on a monthly term?

One H100 SXM 80GB card, 30 days, compute only. The flat number does not move. The hours you put through it do.

Utilisation	Hours used	GPU Rent Hub, flat	Hourly at $2.49	Hourly at $2.99	Hourly at $3.49	GPU Rent Hub per used hour
100%never idle	720	$1,690	$1,793	$2,153	$2,513	$2.35
75%18 hours a day	540	$1,690	$1,345	$1,615	$1,885	$3.13
50%12 hours a day	360	$1,690	$896	$1,076	$1,256	$4.69
25%6 hours a day	180	$1,690	$448	$538	$628	$9.39

Compute only. No egress, storage, image build time or eviction cost on either side. Hourly rates are illustrative of the US on-demand range in 2026, not quotes from a named vendor.

Read that honestly and it says something uncomfortable for us. At 25% utilisation our H100 SXM costs $9.39 for every hour of work you get out of it, against $2.49 on a meter. That is not a rounding error. That is a different product being the right one.

03 Breakeven

At what utilisation does monthly GPU rental become cheaper?

Divide the flat monthly price by the hourly rate. That is the number of hours at which the two are equal.

Compared against	Breakeven, compute only	of 720 h	With $250/mo egress and storage	of 720 h
$2.49/hourcheap end of the US range	679 h	94%	578 h	80%
$2.99/hourmiddle of the range	565 h	78%	482 h	67%
$3.49/hourmajor-cloud end	484 h	67%	413 h	57%

$250/mo assumes 2 TB of outbound at about $0.08/GB and 1 TB of persistent storage at about $0.09/GB-month. Both are included in our price. Neither is included in an hourly compute rate.

The honest conclusion

Against a mid or upper hourly rate, with a normal amount of data moving on and off the card, a monthly term wins somewhere between 55% and 70% utilisation. Below that, hourly wins and you should take it.

Against the cheapest hourly rate, with almost no egress and nothing stored between runs, the crossover moves out to 80–94%. If somebody is quoting you $2.49, a term only pays for itself when the card is close to never idle.

A monthly term is not refunded if you stop on day nine. The hardware is held for you for 30 days whether you use it or not. That is what makes the price flat, and it is also the whole risk of the model.

Utilisation here means GPU-hours you would have paid for on a meter, not SM occupancy. A card left at a shell prompt overnight is billed hourly and costs nothing extra on a term.

04 Hidden costs

What does an hourly GPU price leave out?

The per-hour number is the compute meter. These sit next to it, or never appear at all because they are your time.

Outbound data

Metered clouds bill egress at $0.05 to $0.09 per GB. Two terabytes of checkpoints in a month is $100 to $180 on top of the meter.

Here: 20 TB included, then $4/TB

Persistent storage

The dataset lives somewhere between runs. Block storage at $0.08 per GB-month turns a 1 TB working set into an $80 line that bills while you are not training.

Here: 2 TB local NVMe on the H100 SXM

Idle time while data loads

The meter starts at boot, not at the first training step. Pulling 400 GB and building a container burns 40 to 90 minutes of paid GPU time per cold start.

Here: the data stays on the node

Eviction and restart

Spot rates are the cheap numbers people quote at each other. An eviction costs the gap since the last checkpoint, the requeue wait and the reload.

Here: a term cannot be preempted

Queueing for capacity

On-demand H100 capacity is not always there at the moment you ask. Waiting is not billed, which is exactly why it never appears in a comparison.

Here: 63 H100 SXM free across the three sites

Addresses and ports

Public addresses, guaranteed port speeds and inbound transfer are separate SKUs on most metered platforms. Small numbers, arriving every month.

Here: IPv4, IPv6 /64, unmetered inbound

Our monthly price includes power, cooling, a 100G uplink with a 10G guarantee, one public IPv4 and an IPv6 /64, unmetered inbound, 20 TB of outbound, the local NVMe listed on the GPU catalogue, console access and reinstalls. What is not included is listed on the pricing page.

05 When hourly wins

When should you use an hourly provider instead?

Not a rhetorical section. There are workloads we are the wrong answer for.

If you need a card for six hours, use an hourly provider. We do not sell hours. A monthly term for one afternoon costs $1,690 and returns about $18 of value.

The shape that loses on a term is bursty: a few hours of inference a day, a single benchmark, an evaluation you might abandon on Tuesday, 200 GPUs for 48 hours and then never again. Metered billing exists for that shape.

The shape that wins is boring and continuous. An eleven-day fine-tune. An endpoint that answers at 04:00. A render queue that is never empty. If the card would be sitting there anyway, paying for the hours it sits there is the expensive option.

Take hourly if

You need fewer than about 400 GPU-hours a month, you cannot predict next week's load, you need 50 cards for two days, or you want an hour each on an A100 and an H100 SXM to decide.

Take a term if

The job runs for days, the endpoint is always on, you move real volumes of data, you have been evicted from spot capacity once already, or you want to pay in crypto without an identity check.

06 Every card

What is the per-GPU-hour equivalent of each monthly price?

Monthly divided by 720 hours, weekly divided by 168. Compare against whatever hourly rate you have been quoted for the same card.

GPU	Monthly	Per GPU-hour	Weekly	Per GPU-hour
B200180 GB HBM3e	$3,890/mo	$5.40	$1,090	$6.49
H200 SXM141 GB HBM3e	$2,290/mo	$3.18	$640	$3.81
H100 SXM80 GB HBM3	$1,690/mo	$2.35	$475	$2.83
MI300X192 GB HBM3	$1,590/mo	$2.21	$450	$2.68
H100 PCIe80 GB HBM2e	$1,490/mo	$2.07	$420	$2.50
A100 SXM80 GB HBM2e	$990/mo	$1.38	$280	$1.67
RTX 509032 GB GDDR7	$540/mo	$0.75	$155	$0.92
L40S48 GB GDDR6 ECC	$520/mo	$0.72	$148	$0.88
RTX 6000 Ada48 GB GDDR6 ECC	$460/mo	$0.64	$130	$0.77
RTX 409024 GB GDDR6X	$390/mo	$0.54	$110	$0.65
RTX A600048 GB GDDR6 ECC	$290/mo	$0.40	$82	$0.49
L424 GB, 72 W	$250/mo	$0.35	$70	$0.42
A4048 GB GDDR6 ECC	$240/mo	$0.33	$68	$0.40
RTX 309024 GB GDDR6X	$230/mo	$0.32	$65	$0.39
RTX A500024 GB GDDR6 ECC	$190/mo	$0.26	$55	$0.33

Figures assume 100% utilisation. Divide by your real utilisation to get what you would actually pay per hour of work. Specs and per-region stock are on the GPU catalogue.

07 Terms

Is a weekly GPU rental a sensible middle ground?

Seven days costs about 28% of the monthly price, a 20% premium per GPU-hour, and it converts to a monthly term at the end without reprovisioning.

H100 SXM, one card	Price	Hours in term	Per GPU-hour
1 week	$475	168	$2.83
1 monthauto-renewing	$1,690	720	$2.35
3 months prepaid−5%	$4,817	2,160	$2.23
6 months prepaid−10%	$9,126	4,320	$2.11
8× H100 SXM node, monthly−10% multi-GPU	$12,170	5,760 GPU-h	$2.11

Prepaid discounts stack with the multi-GPU discount. An 8-way node prepaid for six months lands at $1.90 per GPU-hour. Twelve-month terms and full racks are quoted by sales.

A week at $475 beats a $2.99 hourly rate only if you use about 159 of its 168 hours, so weekly is a poor deal for bursty work and a good one for a seven-day run you do not want interrupted. Most people use it as an evaluation before committing to a month, or to hold specific hardware in DFW-1 Dallas, IAD-1 Ashburn or PDX-1 Hillsboro while a project starts.

Payment does not change with the term. Fund a balance once in BTC, ETH, USDT, USDC, XMR, SOL or LTC and renewals are drawn from it 24 hours before term end. No card on file, no invoice address, no identity document. See how crypto payment works and what no-KYC GPU hosting means.

BTCETHUSDTUSDCXMRSOLLTC

08 Questions

Monthly vs hourly GPU rental, asked plainly.

Is monthly GPU rental cheaper than hourly?

Above roughly 55–70% utilisation, yes. A GPU Rent Hub H100 SXM is $1,690 a month flat, which is $2.35 per GPU-hour if the card runs the full 720 hours. US on-demand rates for the same card in 2026 run about $2.49 to $3.49, plus egress and storage. Below about 400 used GPU-hours a month, hourly is cheaper and you should use it.

How many hours a month do I need before monthly is cheaper?

Divide the monthly price by the hourly rate you are comparing against. For our $1,690 H100 SXM that is 679 hours against $2.49, 565 against $2.99 and 484 against $3.49, counting compute alone. Add the $250 or so of monthly egress and storage that metered platforms bill separately and those thresholds fall to 578, 482 and 413 hours.

What is the hourly equivalent of GPU Rent Hub's monthly H100 price?

$2.35 per GPU-hour, being $1,690 divided by the 720 hours in a 30-day term. On a six-month prepaid term it is $2.11, and an 8× H100 SXM node at $12,170 a month is also $2.11 per GPU-hour. Those figures assume the card is busy throughout. At half utilisation the same term works out at $4.69 per hour of real work.

Can I rent a GPU for a few hours?

Not from us. Our shortest term is seven days, $475 for an H100 SXM. If you need a card for six hours, use an hourly provider. We built the business around flat terms on dedicated hardware, which means no meter, no preemption and no queue, and it also means we are the wrong choice for short bursts.

Does the monthly price include bandwidth and storage?

Yes. Every instance includes power, cooling, a 100G uplink with a 10G guarantee, one public IPv4, an IPv6 /64, unmetered inbound, 20 TB of outbound a month and the local NVMe listed for the card, which is 2 TB on the H100 SXM. Outbound beyond 20 TB is $4 per TB. Most invoices have no egress line at all.

What happens if I stop using the GPU halfway through the month?

The term runs to its end and is not refunded, because the hardware is held for you for the full 30 days. You can stop and restart the instance as often as you like at no extra cost. If your balance is empty at renewal the instance stops and the slot is held for 72 hours while you top up.

Do I need to verify my identity to rent a GPU monthly?

No. An account is a username and a password. No email address, no phone number, no identity document, no company details. You fund a balance in BTC, ETH, USDT, USDC, XMR, SOL or LTC and terms are drawn from it. GPU Rent Hub Compute LLC has operated this way from Dallas since 15 March 2021.

More on terms and renewals in the general FAQ and the documentation. If no term fits, you can buy the card outright or colocate hardware you already own.

Run the numbers against your own utilisation.

The pricing calculator uses the same math as checkout. Minimum deposit $20 equivalent, balances never expire, withdraw what you do not use.

Create an account →