Run the numbers before you buy
Stop renting GPUs by the hour
Rent vs. own over 3 years
24 hours a day
- Ultra 9 · RTX PRO 6000$28,799
- AWS on-demand$3.36/hr
- GPU cloud median$2.36/hr
Move across the graph to see any month
How we worked this out
1 GPU, on-demand prices per GPU-hour. Ultra 9 · RTX PRO 6000: $28,799 once, plus power at 950 W (estimated) and $0.18 per kWh. Cloud prices as of November 2026: AWS g7e.2xlarge on-demand, and the median on-demand price across the providers GetDeploying tracks (48 for the RTX PRO 6000, 19 for the RTX 5090). Power is estimated from rated part power: the GPUs and CPU at their maximum, plus 100 W for the rest of the system.
No uploads. No waiting.
Time before you can start
Shorter is better
- Upload to the cloud at 100 Mbps 22 hours
- Upload to the cloud at 1 Gbps 2 h 13 min
- Read from your Novatech's SSDPCIe 5.0 NVMe, rated at 14,700 MB/s 1 min 8 s
How we worked this out
Best case: the full line speed with no overhead, so real uploads take longer. The SSD time uses the rated read speed of the 2 TB PCIe 5.0 drive in every workstation.
Fine-tune on your own data
GPU memory by build
- RTX 5070 Ti workstation16 GB GPU memory · from $3,999 Too small
- RTX 5090 workstation32 GB GPU memory · from $11,499 Fits
- RTX PRO 6000 workstation96 GB GPU memory · from $28,799 Fits
- 2× RTX PRO 6000 server192 GB GPU memory · from $92,998 Fits
How we worked this out
Minimum GPU memory from Unsloth's published fine-tuning requirements. Longer context and bigger batches need more. On the dual-GPU server the model is split across both cards.
Your AI systems expert
- POs and purchase cards welcome
- Net terms
- University & government procurement
WHAT HAPPENS NEXT
-
An expert reads it.
Your models, budget, and preferences, reviewed by the person who would build the system.
-
You get a real spec.
Configuration with pricing you can hand to procurement, in one business day.
-
You decide.
No obligation. If a smaller build serves your models better, that is what we recommend.
