AI compute that lives in houses — not another hyperscale shed.
Hestia Grid puts owned GPU clusters inside residential property: sell sovereign AI compute, collect rent, and put surplus heat and solar to work on-site. The model travels; we prove it first where it’s most viable — France — then grow with demand.
One asset class. Three useful outputs.
Sovereign AI compute
GPU hours sold from real buildings you can point to — EU residency when you need it, outside the hyperscale queue.
Lived-in property
The house still houses people. Rent (or lease) is part of the stack, not an afterthought.
Energy that would otherwise waste
GPUs throw off more heat than one home can use; roofs can throw off more solar than the site needs. We capture what we can, dump or export the rest — honestly.
Customers on both sides of the door
AI compute buyers
Labs, startups, and teams that need GPU time — training, inference, batch — on capacity that isn’t locked inside a US cloud campus. That’s the primary revenue line.
Tenants & leaseholders
People who live in (or lease) the property. They get a normal home; heating bills can be partly offset when waste heat feeds the house in winter.
Site capital (later)
Partners who want to own individual house-nodes while we originate and operate them. Not this week’s raise — the scale path after a live site.
More heat and solar than the house alone needs
An 8-GPU node produces far more annual heat than one dwelling can absorb — most of it is rejected in summer; winter reuse is a bill amenity, not the reason to buy GPUs. On-site solar can likewise oversupply the plot at midday. The design uses what the building can take, meters the rest, and keeps compute + rent as the economics that matter.
Operate the network. Own sites with partners.
Hestia runs origination, install, and GPU sales. As sites prove out, partners can own house-nodes one SPV at a time — club deals first, a formal vehicle only after audited track record. Returns are meant to come from compute hours and rent, not from heat or carbon coupons.
Three steps, no commitment required yet
- 1
We size the first cluster
Your reservation shows real demand — where, how much, when — before we lock capital into a build.
- 2
You choose your level
Join the waitlist for updates, or submit a formal non-binding Letter of Interest if you want to anchor early capacity.
- 3
We come back with terms
When the first site is confirmed, registered parties hear first — pricing, capacity, and go-live.
Tell us what you'd need
Two minutes. No payment, no obligation.
Indicative pricing: €1.35–€2.40 per GPU-hour at ~70% utilisation on an 8-GPU node. Confirmed rates shared before go-live.