Torus raised $500M to drop AI compute containers behind existing meters. Electrical subs should ask who does the week of site prep

Torus, a Utah energy-storage manufacturer, raised $500 million on November 11 to build Foundry: factory-made AI compute units of 1 to 20 MW that sit behind the meter at existing commercial and industrial sites. For an electrical or sitework sub, this is data center work at the scale of a service upgrade, not a campus. Torus says each site needs about a week of prep and no building, and it has not said who does that week.

What did Torus announce?

Per the company's release and MLQ's coverage of the round, which Bloomberg reported first:

Every number in that table is a company claim. Torus names one customer, GPU cloud firm Hydra Host, and says "thousands" of GPUs are deployed without giving a count or a site.

Why does this matter to an electrical sub?

Most of the AI construction we cover is gigawatt campuses that a regional sub only sees as a third-tier package. Foundry points the other way. CEO Nate Walkingshaw put it as "Foundry brings the chips to power," meaning the unit goes wherever a plant or warehouse already has unused transformer capacity.

If that model works, the work is many small jobs: a pad or prepared surface, feeders and protection from an existing service, a fiber path, grounding, and commissioning support. That is ordinary scope for a commercial electrical contractor with a sitework partner. We made a similar point about air-cooled AI racks going into existing rooms.

The catch is that Torus sells this as "full stack, one provider" and turnkey, with its own operations and maintenance. The release does not say whether site prep is self-performed, subbed locally, or left to the host.

What does the announcement leave out?

• Utility review. Torus claims no new grid infrastructure and no reliability impact. It does not mention utility approval or interconnection study time.
• Permitting. Nothing on building, electrical or fire review for a unit that combines battery storage with megawatt-class loads.
• What is in the 12 to 16 weeks. It is not stated whether permits and utility sign-off fall inside that window.
• Storage size. No MW, MWh or backup duration for the paired storage.
• Price and contract terms. None published.

What should a sub ask before pricing one of these?

1. Who holds the site-prep contract? Torus, the host, or the compute customer. That decides whose payment terms you live under.
2. Who files for permits, and under which codes? Ask how the local authority is treating the battery and flywheel storage and what separation or fire protection it wants.
3. Who proved the transformer headroom? Get the load study and the utility's written position, not a sales estimate.
4. Where is the scope line? Find the exact point where your conductors stop and factory wiring starts, and who owns testing across it.
5. What happens if the unit ships late? A one-week prep window means your crew is scheduled against a factory delivery date.

Takeaway: Do not chase this yet. One named customer and no named sites is thin. But if an industrial client asks whether their spare service capacity could host compute, you now know what the pitch looks like and the five answers to get in writing before you price it. The earlier modular push we covered, Crusoe's factory-built AI modules, raised the same question about how much field labor is left.

Construction AI Brief publishes three times a week. Subscribe at constructionaibrief.com and forward this to whoever prices your service and distribution work.