Webinar | Thursday, Dec. 3, 2026 | 11 a.m. PT
What AI training infrastructure really costs
The lowest GPU-hour rate isn't always the lowest bill.
Join Surf Report and Crusoe's solutions engineering team to see what drives AI training infrastructure’s TCO & ROI and how to model your own with best practices.
Speakers:
Jake Dolezal
Distinguished Engineer, AI, Data & Infrastructure
Surf Report
Surf Report
Rohan Mathews
Senior Director, Solutions Engineering
Crusoe
Crusoe
Michael YenChi Ho
Director, Technical Product Marketing
Crusoe
Crusoe
The hourly rate is the easiest number to compare across AI infrastructure providers. It also leaves out what it takes to run the platform: the engineers, the time to your first training step, and how much of your spend becomes finished work when something fails.
Surf Report tested that full lifecycle. Its engineers ran comparable training workloads on Crusoe Cloud, a competing neocloud and a hyperscaler, injected failures and modeled three years of cost for a lean team. The lowest quoted rate didn't deliver the lowest total cost.
In this session, one of the engineers behind the study walks through what they tested, what they found and where the results stop.
Surf Report tested that full lifecycle. Its engineers ran comparable training workloads on Crusoe Cloud, a competing neocloud and a hyperscaler, injected failures and modeled three years of cost for a lean team. The lowest quoted rate didn't deliver the lowest total cost.
In this session, one of the engineers behind the study walks through what they tested, what they found and where the results stop.
What you’ll learn:
- How compute, staffing and reliability add up to three-year TCO & ROI
- Why cost per token and Goodput tell you more than the hourly rate or peak throughput
- What to check in your first week on any platform
- Four questions to ask any provider before you sign
From the study:
| Metric | Crusoe Cloud | Neocloud | Hyperscaler |
|---|---|---|---|
| Three-year TCO | $5.36M | $6.14M | $8.37M |
| Engineers to run the platform | 3.9 | 4.1 | 6.3 |
| Cost per 1M training tokens | $2.89 | $3.32 | $4.39 |
| Goodput under injected failures | 99.1% | 94.2% | 95.9% |
Crusoe Cloud's three-year TCO was 36% lower than the hyperscaler's, a $3.01M difference. Its 99.1% Goodput provided 12 to 18 additional full days of productive training per year.
The True Engineering Cost of Open Model Training: A TCO and ROI Analysis of GPU Infrastructure. Independent testing and analysis by William McKnight and Jake Dolezal of Surf Report. Sponsored by Crusoe. Results reflect lab testing of training workloads, published in September 2026. Production results will vary.