GPU cloud bill audit
Enter your GPU cloud bill and see what owning the same GPUs would cost over three years.
Your bill
Fine-tune
Owning saves, over 3 years
$812K
Your 64 H200 GPUs against owning 8 H200 servers, after resale and your cloud bill for the 3 months until the servers are live.
- Your GPU bill a month
- $145K
- Owning pulls ahead
- Month 35
- Per GPU-hour used, 3-yr average, renting vs owning
- $3.82 vs $3.13
Your rate against the market
You pay $4.44 per H200 GPU-hour, the same as the median on-demand price across 46 providers ($4.44).
On-demand H200 prices, $2.09 to $10.60 across 46 providers, checked October 17, 2026. Each provider’s lowest listed price, from its own pricing page. See the GPU cloud price comparison.
Cheapest: Beam $2.09, Airon AI $2.63, HPC AI $2.68, DeepInfra $2.69, Bitdeer $2.76.
See every line
Renting, as on your bill
- 64 H200 GPUs, on demand
- $4.44 per GPU-hour
- Bill in a 730-hour month
- $145K
- Total over 3 years
- $4.50M
Owning
- 8 H200 servers, network, shipping, setup
- From $3.04M
- Your cloud bill for the 3 months until it’s live
- $431K
- Data center and management
- $889K
- Resale value (25%)
- -$680K
- Total over 3 years
- $3.68M
Ballpark, before taxes, insurance, and financing. Owning takes cash upfront, or about half with a lender (up to 70% with a signed customer contract), and 60-90 days to go live, while your cloud bill keeps running.
Rent or own: four common GPU bills
64 GPUs in use 70% of the time, on demand at the median price across the providers in our GPU cloud price comparison, or reserved at the 3-year contract rate we see in deals, against owning the same number of GPUs as 8-GPU servers for 3 years. H100 bills compare with H200 servers, the Hopper server sold new today. Idle hours are not rented out.
| Bill | Per GPU-hour | A month | Renting, 3 yrs | Owning, 3 yrs | Owning saves | Owning ahead |
|---|---|---|---|---|---|---|
| 64 H200, on demand at the market median | $4.44 | $145K | $4.5M | $3.7M | $812K | month 35 |
| 64 H100, on demand at the market median | $3.49 | $114K | $3.5M | $3.6M | -$55K | not within 3 years |
| 64 B200, on demand at the market median | $6.82 | $223K | $6.9M | $5.1M | $1.8M | month 30 |
| 64 B300, reserved 3 years | $4.60 | $215K | $7.7M | $5.7M | $2.0M | month 31 |
Renting: median on-demand price per GPU-hour across the providers in our comparison, checked October 17, 2026. Owning: our ballpark server prices (October 2026), networking, shipping, our fees, data center rent, the cloud bill for the 3 months until the servers are live, and 25% resale. Before taxes, insurance, and financing.
When owning beats your cloud bill
Steady use for two years or more. A reserved bill pays for every hour whether you use it or not. On demand at 70% use, 64 H200 GPUs at the market median work out to $3.82 per GPU-hour used over 3 years, against $3.13 owned. Servers you own still sell for something at the end. The more you pay above the median, and the more of the time you use the GPUs, the more owning saves.
Keep renting if your usage is spiky, you need the GPUs for a few months, or you can’t put cash or a down payment into hardware. Owning also takes 60-90 days to go live, and the audit counts data center rent from the day you order and your cloud bill until the servers are live.
Idle hours don’t have to be wasted. When you own, the hours you don’t use can be rented to other AI companies. The audit leaves that out unless you turn it on, and then counts it as optional upside: only the GPUs you pay for today, at the market median, 52% of the time. See how renting out GPU servers works.
How the audit works
Updated October 17, 2026. Cloud prices checked October 17, 2026, across 117 providers; server prices October 2026.
Your numbers. Effective rate is GPU spend divided by GPU-hours. A bill is scaled to a 730-hour month. On demand, the GPUs you need to own are the most that run at once; reserved, the GPUs reserved. Owning rounds up to whole 8-GPU servers.
The comparison. The same math as our rent vs own calculator: reserved rent fixed for 3 years, on-demand rates falling 10% a year, data center rent from month 1, 3 months to go live with your cloud bill still running (counted on the owning side), our flat fees (no commission on hardware or the data center), and resale at the end. Break-even is the first month owning has cost less in total, before resale.
Market rates. On-demand prices per GPU-hour across 117 GPU clouds, checked October 17, 2026, each provider counted once at its lowest listed on-demand price for the GPU, read off its own pricing page. Hyperscaler rates are the instance price divided by its GPUs. The full table, with sources, is the GPU cloud price comparison.
Talk to us about owning.
- A call with our CTO about what you need
- Ballpark costs and timelines for your budget
- A straight answer if it’s not a fit
Prefer email? hello@amcompute.com
GPU bill audit questions
- Should an AI startup buy or rent GPUs?
- Buy when you use the GPUs most of the time for two years or more; rent when usage is spiky, short, or you can’t tie up the cash. 64 H200 GPUs on demand at the $4.44 market median, in use 70% of the time, cost about $4.5M over 3 years; owning 8 H200 servers costs about $3.7M, counting your cloud bill until the servers are live and resale at the end. At the H100 median, renting costs about $55K less than owning 8 H200 servers.
- How is my effective rate figured?
- GPU spend divided by GPU-hours billed. Hyperscalers bill per instance, so an 8-GPU instance-hour is eight GPU-hours. Storage, networking, and support are left out.
- What does the owning side include?
- The servers, cluster networking, shipping, our $25K setup fee, data center rent at $280 per kW a month, management from $1K a month per 4 servers, your cloud bill for the 3 months until the servers are live, and 25% resale after 3 years. Not taxes, insurance, or financing.
- Does the audit count renting out idle hours?
- Only if you turn it on, and it is shown as optional upside. It counts the unused hours of the GPUs your bill pays for today (not spare GPUs in a rounded-up server), at the median on-demand price across the providers we track, rented 52% of the time. If owning only wins because of that income, the audit says so.
Further reading
Other tools
- GPU earnings calculatorWhat owning GPU servers can earn, with ballpark costs and real rental rates.
- Rent vs own calculatorCloud GPU rent next to the cost of owning the servers, for startups.
- Bonus depreciation plannerFederal depreciation on GPU servers by tax year with 2026 bonus depreciation, the tax it defers at your bracket as an estimate, the servers’ pre-tax earnings, and the passive activity rules. General information, not tax advice.
- MACRS depreciation calculatorYear-by-year tax depreciation for GPU servers from the IRS MACRS tables, with bonus depreciation and section 179 amounts you enter.
- Colocation pricing by metroColocation asking rates per kW a month by North American metro and deal size, with power costs and signed rents, each from a dated public source.