Trace builds and manages inference sites for secure, low-latency GPU compute.
A site, in section
Trace deploys GPU inference inside commercial buildings, using the power, cooling, and fiber they already have.
The plant room
The racks go into a plant room or a similar service space that the building already has.
The problem
A new data center takes years to open, because it has to wait for a power connection, for permits, and for construction. Meanwhile, many office buildings use far less electricity than they were built for, and they already have cooling and fiber in place.
The idea
How one office building becomes a place to run inference
Trace installs and operates the hardware in capacity the building is not using, pays the owner for hosting it, and sells the compute to AI teams nearby.
Why it matters
Why the distance to the GPU adds up
An AI agent does not make one request. It makes hundreds, one after another, and every one of them travels to the GPU and back. A faster model shortens the thinking, but it does nothing for the trip.
Take a task of 1,000 calls made in sequence. If each round trip takes 70 milliseconds, the task spends 70 seconds on the network before any computing is counted. If each round trip takes 3 milliseconds, it spends 3 seconds.
This is arithmetic, not a measurement: 1,000 calls multiplied by the round-trip time. Ask us for measured figures from the site nearest your users.
Two ways to work with Trace

You own buildings with electrical capacity that nobody is using, and you would like it to earn something.
Where we are
Ten cities across North America
- CitiesTrace operates in Montreal, New York City, Houston, and seven other major cities in North America.
- SpeedA new site goes from signed agreement to serving customers in less than 90 days.
- FundingTrace Compute has raised $250M+. We're now building a network of inference sites.
- HomeThe company is based in New York.
