How much power can a colocation site sell?
Choose a customer mix and a risk level. The estimate gives the largest contracted book the installed IT capacity can carry, as a ratio of built power, with the risk of exceeding it. Every default is an illustrative example; change any of them.
Customer list
One row per contract. Type over any cell, or paste rows from a spreadsheet. Contract sizes set the concentration: one large tenant and many small ones behave like a few customers, and the estimate counts them that way.
Peak use is the share of its contract a customer draws in its busiest 1% of hours. Leave it blank to use its segment's behaviour, shown in grey.
Saved lists
Sign in to save this list to your account. Saved lists are kept on our server under your account only, and deleted with it.
The list stays in this browser tab unless you save it to your account; closing the tab clears an unsaved list.
Today's book against the capacity
Contracted power, expected draw and the draw not exceeded at the chosen risk level, on one scale with the installed and the tested capacity.
By segment
Today's book split by segment: what each draws on average, how much of the hourly swing it brings and the correlation used.
Deeper views
How the estimate works
Each segment's customers draw a mean share of their contracted kW with an hourly swing. Customers of one segment move together by the correlation shown; customers of different segments by the correlation between segments.
The draw of the whole book is the sum. Its mean and swing scale with the book, so the largest book is the capacity tested against divided by the draw per contracted kW at the chosen risk level. The ratio is that book over the installed capacity, and it can fall below 1.
A few large customers behave like a few customers. With the customer list, each segment counts its customers by their contract sizes (the square of the total over the sum of squares), so one large tenant weighs as it should; a customer's peak use scales its segment's behaviour to that peak. When the mix is entered as shares, the estimate counts the customers per segment shown in the table.
Not modelled: loading per feeder or per PDU, redundancy topologies beyond one headroom share, protection coordination and long-run growth of each customer into its contract.
Read what capacity sells for in each market Open the market pages
Sources: NFPA 70, National Electrical Code, 210.20(A); IEC 60947-2:2024, low-voltage circuit-breakers; IEC Electropedia 691-10-03 and 691-10-04; Fan, Weber and Barroso, Power provisioning for a warehouse-sized computer, ISCA 2007; Kumbhare et al., Prediction-based power oversubscription in cloud platforms, USENIX ATC 2021; Patel et al., Characterizing power management opportunities for LLMs in the cloud, ASPLOS 2024; Power stabilization for AI training datacenters, arXiv 2508.14318, 2025.
An estimate on illustrative values, dated, for screening a sales policy. It is not advice and makes no recommendation on what to sell.
ScienceShot calculation.
Cite this page
ScienceShot. (2026). Oversell estimate [Online calculator]. https://scienceshot.com/sell/oversell
@misc{scienceshot-oversell-2026,
author = {{ScienceShot}},
title = {{Oversell estimate}},
year = {2026},
note = {Online calculator},
url = {https://scienceshot.com/sell/oversell}
}