Energy & grid access
Electricity, interconnection and grid capacity
Industry map / supply chain
GPU rental prices are the last step in a longer chain. Utilities make power available; contractors and equipment makers help build facilities; data-centre operators provide critical IT capacity; GPU hosters sell compute. Each participant earns a different kind of revenue and carries different capital risk.
01 / Economic chain
Electricity, interconnection and grid capacity
Transmission, electrical works and construction services
Switchgear, UPS, cooling and other facility equipment
Land, buildings, critical IT power and high-density hosting
Accelerators, servers, memory and cluster networking
GPU instances and computing services to AI customers
These are industry-role examples, not a claim that every named upstream company supplies every GPU hoster. Named customer relationships appear below only where a source identifies both parties.
02 / Contracts with identified counterparties
High-density colocation and leased customer power · Multi-site CoreWeave hosting agreements
Approximately 590 MW leased power capacity
Contracted/leased; not a billable-capacity figure
Core Scientific owns and develops hosting infrastructure; CoreWeave uses that capacity for compute. Separate landlord cost and rent from GPU systems and GPU-hour revenue.
2025 Form 10-K ↗ · Item 1, High-Density Colocation business and CoreWeave contracts · reporting basis 2025-12-31
Long-term data-centre leases and critical IT load · Polaris Forge 1
400 MW contracted critical IT load across three leases
Contracted; phases have different operational and delivery dates
Track Applied Digital's facility return and CoreWeave's compute return separately. The same 400 MW is capacity supplied to one hoster, not 800 MW of industry capacity.
FY2026 Form 10-K ↗ · Business overview, Polaris Forge 1; contracted lease portfolio · reporting basis 2026-05-31
03 / What we will measure
| Supplier layer | What reaches the hoster | Model cost line | Research signal |
|---|---|---|---|
| Energy & grid access | Electricity, interconnection and grid capacity | Energy, demand charges, connection and minimum commitments | Tariffs, large-load agreements, connection contributions and delivery dates |
| Grid & site construction | Transmission, electrical works and construction services | Project capital and delivery schedule | Contract scope, backlog conversion, lead times and cost overruns |
| Power & cooling systems | Switchgear, UPS, cooling and other facility equipment | Facility capital and operating efficiency | Orders, shipments, equipment mix and supply constraints |
| Data-centre capacity | Land, buildings, critical IT power and high-density hosting | Facility capital or contracted hosting rent; depends on asset ownership | Built, contracted, online and billable IT MW by site and phase |
| GPU & server systems | Accelerators, servers, memory and cluster networking | Installed system capital, refresh and hardware availability | Hardware shipments, generation mix, lead times and useful lives |
| GPU hosting | GPU instances and computing services to AI customers | GPU systems, facility payments, power and support | Posted versus realized GPU-hour price and paid utilization |
Supplier backlog, customer capex and hoster revenue can describe different stages of the same investment. The project, payer, asset owner, cost scope and date must match before we derive cost per MW or industry totals.