Most buildings use only part of the electrical capacity already wired to them. LoadShare places AI computers in that unused room, watches the transformer every second, and switches the computing off whenever the building or the grid needs the power back. The building owner earns a monthly hosting payment.
Free screening estimate · No equipment is installed until a site study and your utility confirm it is safe
Computing is the most flexible load on the wire. LoadShare treats it that way.
Your address, building and (optionally) power bill give a year of 15-minute load against real local weather. We find how much room the transformer has, hour by hour.
Before anything is installed, the utility confirms the transformer and service. Nothing goes in until the site study shows it is safe.
A site controller measures the load and lowers or switches off the computers when the building peaks, when the grid calls a demand-response event, or if anything looks wrong.
The AI operator pays for its own metered power plus a monthly hosting fee per computer, and the work moves to other sites whenever this one needs its capacity back.
The same parts at every site: the computer in a sealed outdoor enclosure, its own disconnect and meter, and a site controller that can always switch it off.
Enter the street address. We look up the building and its local weather and estimate what it could host. It takes about half a minute.
Simulated on public NREL ResStock and ComStock building data for Mecklenburg County, NC, with 2018 weather, in nameplate mode: the transformer is never loaded past its rating.
8 homes on one 37.5 kVA pole transformer
36 units in 3 buildings on a 112.5 kVA pad-mount
75,000 sq ft on a 750 kVA transformer
46,000 sq ft on a 300 kVA transformer
300,000 sq ft on a 1,500 kVA transformer
21,000 sq ft, two shifts, 225 kVA
Without control, the same computers would push these transformers to 166% of their rating at the worst hour. Owner income shown at an illustrative $25 per computer per month; actual terms depend on the site and the operator. Read the background in Homes as grid-aware AI infrastructure.
No. The computers are on their own meter, and the AI operator pays for that power. The owner’s bill does not change.
The site controller keeps the total load at or under the transformer’s rating, using its own current sensors, and a hardware contactor cuts the computing feeder if the software ever fails to. The utility reviews the site before anything is installed.
The computing is the first load to go. In the examples, every one of the year’s demand-response events was honored by shutting the computers off for the event.
It is an estimate from public data with a 10% safety margin on the building load, checked against real reference buildings. A power bill makes it better; a year of 15-minute meter data from your utility makes it much better. A site study comes before any installation.
Every screening run is recorded in our model journal (address, building facts and results) so we can show how the program performs across many sites. Names and emails are kept separately and only used to send your report and follow up. If you add a bill or meter file, we save only the usage figures, utility and ZIP, and do not keep the file. See the privacy policy.
We are looking for a small number of pilot sites and utility partners. The site study is free.
Talk to usBuilding outlines © OpenStreetMap contributors (ODbL). Weather: NASA POWER. Addresses: US Census Geocoder. Reference buildings: NREL ResStock and ComStock. Estimates only; not a utility interconnection study.