# Land, Power, and Shell

> Source: https://aiwiki.ai/wiki/land_power_and_shell
> Updated: 2026-08-20
> Fact-checked: 2026-08-20
> Categories: AI Infrastructure, Data Centers
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Land, Power, and Shell." aiwiki.ai, 20 Aug 2026. https://aiwiki.ai/wiki/land_power_and_shell
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Land, power, and shell** (**LPS**) is an emerging commercial label for the physical site, deliverable electricity, and building structure needed before computing equipment can be installed in a large [data center](https://aiwiki.ai/wiki/data_center). The phrase is used in the development and financing of AI factories, but it does not have a universal technical specification. Its components overlap with the older real-estate categories of powered land and powered shell, whose scope varies by project and contract.[1][2][5][6][7]

LPS usually stops short of a fully fitted, compute-ready facility. Servers, [GPUs](https://aiwiki.ai/wiki/gpu), storage, network equipment, and software are normally outside the category. Cooling, uninterruptible power supplies, internal electrical distribution, and other mechanical, electrical, and plumbing systems may be delivered by the shell owner in one transaction and left to the tenant in another. The lease, development agreement, technical schedules, and acceptance tests therefore matter more than the label alone.[5][6][7][13]

## Development stages

LPS sits within a development continuum rather than a single standardized package. Industry sources use related terms for stages that transfer different combinations of land risk, utility risk, construction work, and operating responsibility.[5][6][7][8][9]

| Stage | Typical physical and commercial scope | Usually still required |
|---|---|---|
| Development land | Site ownership or control, preliminary planning, and possible zoning or permitting work. Power may be only proposed or under study. | Firm power rights, utility works, building construction, and fit-out.[8][9] |
| Powered land | Entitled or developable land with a contractual route to a stated amount of utility power, often for a specified delivery period. Some offerings also include site preparation or fiber planning. | Building shell, interior systems, commissioning, and computing equipment.[5][8][9] |
| Powered shell | Structural envelope and agreed core infrastructure, often including exterior improvements and a defined power and fiber handoff. | Tenant-specific mechanical and electrical fit-out, data halls, racks, and computing equipment to the extent excluded by contract.[5][6][7] |
| Shell plus fit-out or turnkey facility | Building plus some or all cooling, internal power distribution, controls, telecommunications, security, and data-hall systems. | Computing equipment and software, unless separately included.[5][7] |
| Compute-ready facility or operated capacity | Commissioned facility with racks, servers or accelerators, networking, storage, and an operator or service arrangement. | Workloads, applications, and any customer-specific services. This stage is generally beyond LPS.[2][5] |

These categories can overlap. A build-to-suit lease describes how a facility is procured for a specific user, while powered shell describes a delivery condition. A build-to-suit project can end at powered shell, at a fitted building, or at a commissioned facility. Wholesale [cloud computing](https://aiwiki.ai/wiki/cloud_computing) or colocation agreements can also combine real-estate rent, reserved power, metered energy, cooling charges, cross-connects, and operating services in different ways.[6][7]

## Land and site control

The land component includes more than acreage. A usable site needs legal control through ownership, a ground lease, or another enforceable right; appropriate zoning and permits; physical access; space for buildings and electrical yards; and routes for fiber, transmission, and utility infrastructure. Title restrictions, easements, environmental conditions, construction logistics, and the ability to subdivide or expand a campus can affect both delivery and later financing.[8][9]

Water and cooling feasibility are also site-specific. Water availability depends on the local watershed, climate, competition with other users, and the selected cooling design. The choice between water-cooled and air-cooled systems changes the facility's water-use profile and should reflect local conditions. The word "land" in LPS does not establish that these questions have been resolved.[12]

Powered-land transactions move some risk from a future buyer or tenant to a land developer. The developer may obtain entitlements and a contractual right to power, then sell or lease a site that is closer to construction. This can avoid operating and tenant risks for the powered-land investor, but it also leaves the building and long-term operating cash flow to another party.[8][9]

## Power rights and deliverability

Power is both a legal commitment and a physical delivery system. A nearby substation, transmission line, or generating plant does not by itself make a parcel powered. Diligence normally examines the counterparty, contracted capacity, delivery date, interconnection studies, required network upgrades, construction responsibility, security deposits, tariff or power-purchase terms, curtailment rights, redundancy, and remedies if service is delayed or unavailable.[9][10]

Several milestones should remain separate in project descriptions: a utility application, an interconnection position, an executed agreement, completed network upgrades, energization, facility commissioning, and power that passes the tenant's acceptance tests. A project can have a contractual right to future capacity while still carrying schedule and construction risk. Legal analysis of powered land treats the contractual right as characteristic of the opportunity while warning that the strength and timing of that right vary materially.[9]

Large data-center loads also interact with grid planning and regulation. In the United States, the Federal Energy Regulatory Commission's RM26-4 advance notice asked how loads generally above 20 MW should enter the interstate transmission system, who should pay for upgrades, and whether flexible loads should receive a faster study process. On June 18, 2026, the Commission used that record to issue separate show-cause orders directing six regional grid operators to justify or reform their tariffs, rather than issuing a proposed final rule. The process illustrates why a power commitment can depend on studies, reliability requirements, cost allocation, and state or federal jurisdiction.[10][17]

Facility power and usable IT power are not interchangeable. Cooling, pumps, lighting, conversion equipment, and other infrastructure consume part of the electricity delivered to a site. AI workloads can also create rapid and large changes in demand. National-laboratory research identifies computational scheduling, facility controls, storage, and on-site generation as possible sources of flexibility, while noting that grid constraints can delay the construction of computing capacity. These issues connect LPS development to the broader subject of [AI energy consumption](https://aiwiki.ai/wiki/ai_energy_consumption).[11][16]

## Shell and fit-out boundary

A powered shell commonly includes the completed building exterior and a contractual handoff for power and network access. One current industry description includes a switchyard or on-site substation, fiber entry points, security, and exterior improvements. Another leaves racks, cooling, and internal electrical systems to the customer. A 2021 investment report similarly separated powered shell and core from mechanical and electrical fit-out and from facility operations.[5][6][7]

The boundary can be narrower than the phrase suggests. In one 2025 ground lease for a proposed data-center campus, the defined tenant-responsibility equipment included uninterruptible power supplies, air handlers, cooling towers, chillers, switchgear, customer-side transformers, generators, servers, racks, networking, cabling, and internal power distribution. That list describes one agreement, not every powered shell. It demonstrates why a technical responsibility matrix is necessary.[13]

Fit-out specifications determine whether a shell can support the intended rack density, cooling method, redundancy level, and hardware generation. They also determine who bears the cost of redesign, long-lead equipment, commissioning failures, and future upgrades. A shell that can accept one tenant's design is not automatically interchangeable with another tenant's facility without additional work.[5][6][8]

## Commercial structures and counterparties

An LPS project can involve a landowner, developer, electric utility, grid operator, generation owner, landlord, colocation provider, architect, engineer, construction contractor, equipment supplier, tenant, operator, lender, infrastructure investor, insurer, and guarantor. [Nvidia](https://aiwiki.ai/wiki/nvidia)'s Omniverse DSX blueprint treats LPS builders and operators as a separate participant group from design and construction firms, power and cooling vendors, systems integrators, and software vendors. That separation is useful because the parties can carry different completion and operating obligations.[2]

Common structures include a sale of powered land, a long ground lease, a build-to-suit lease, a powered-shell lease, and a fully fitted or operated-capacity agreement. Powered-land sellers can realize value after entitlements and power rights are secured. Build-to-suit developers generally construct to a named tenant's specifications under a long lease. Speculative developers start without a committed tenant and retain greater demand, design, and financing risk.[6][8]

The contract should identify the premises and capacity, the point at which power is measured, delivery and ramp dates, completion tests, who obtains permits, who funds utility upgrades, and which party owns each layer of infrastructure. It should also allocate operating costs, taxes, insurance, power-price changes, outages, force majeure, expansion rights, restoration obligations, and end-of-term removal or reuse. Marketing descriptions cannot replace those schedules.[6][9][13]

## Financing and credit support

LPS assets require substantial capital before a tenant can install compute. Project lenders and investors therefore examine construction progress, utility delivery, the tenant's lease obligations, the tenant or guarantor's credit, and whether another user could occupy the site after a default. A long lease can support financing, but its value depends on enforceability, counterparty credit, completion conditions, and the cost and time needed to relet the facility.[8][14][15]

Credit support can take several forms. A parent or affiliate can guarantee a special-purpose tenant's obligations. A tenant can post a letter of credit, cash deposit, or escrow. A developer can provide completion support, while another commercial participant can guarantee specified lease or power payments. One 2026 filing described springing parent guarantees and a $50 million letter of credit for two data-center leases. A separate filed financing agreement defined an acceptable powered-shell customer by reference to net worth, credit ratings, an affiliate guarantee, or lender approval. Those thresholds were deal-specific examples, not market rules.[14][15]

Nvidia disclosed another structure in its fiscal 2026 Form 10-K. The company had guaranteed partners' facility lease obligations in exchange for warrants. The filing reported maximum gross exposure of $3.5 billion, terms of five to seven years, and $712 million placed in escrow by the partners. It classified the guarantees as credit derivatives and said exposure declined as partners paid the lessors. The disclosure shows how a vendor's balance sheet can support LPS leases, but it does not imply that Nvidia funded the facilities upfront or that guarantees are required in every LPS transaction.[3]

### Residual-value guarantees

A residual-value guarantee is one possible credit tool. In the specific agreements Nvidia filed for the [PORTS Technology Campus](https://aiwiki.ai/wiki/ports_technology_campus) on August 17, 2026, Nvidia agreed to cover a defined shortfall between a guaranteed minimum lease value and amounts recovered through reletting or a sale after specified OpenAI insolvency or payment-default events. Nvidia could instead assume the applicable lease, direct a reletting or sale process, permit termination, or delay those remedies for up to one year while paying specified project costs.[4]

The filed agreements relate to leases supporting about 4.25 GW of initial IT load and have a cumulative payment cap of $105 billion. An agreement generally becomes effective only when its lease commences, and payment obligations are subject to conditions that include ready-for-service requirements expected to begin in 2028. The obligations can continue as long as 20 years but may terminate earlier under specified conditions, and OpenAI agreed to reimburse Nvidia for amounts it actually pays. The cap is contingent exposure, not an upfront payment, a valuation of the site, or a prediction of loss.[4]

PORTS is unusually large and has transaction-specific triggers and remedies. It should not define LPS generally. The term "residual-value guarantee" alone does not show whether another agreement concerns property, equipment, lease payments, sale proceeds, a particular period, or defined default events. Any description needs to identify the guaranteed obligation, beneficiary, trigger, recovery process, cap, duration, collateral, reimbursement rights, and accounting treatment.[3][4]

## Nvidia's LPS framing

Powered land and powered shell existed as development and investment models before Nvidia grouped the three words into the LPS label. PGIM Real Estate described powered shell and core in 2021, and later industry material continued to distinguish it from fit-out and operation. Nvidia's fiscal 2026 annual report used the phrase "land, power, and shell guarantees," its DSX blueprint identified LPS builders and operators, and its August 17, 2026 blog called LPS a strategic resource for AI factories.[1][2][3][7]

Nvidia's framing places the real-estate and energy layers alongside chips, networking, systems, and software in a broader [AI infrastructure](https://aiwiki.ai/wiki/ai_infrastructure) supply chain. It can help distinguish a place to install compute from the compute equipment itself. Nvidia also said that most customers would continue to secure their own LPS and that its direct support would be selective. The company's strategy is therefore evidence of one expanding commercial role, not proof that all AI-factory development now follows the same model.[1][2]

## Risks and due diligence

The value of LPS depends on linked rights and delivery obligations. A failure in one layer can strand investment in another. Important diligence areas include:[8][9][10][11][12][13]

- **Site and entitlement risk:** confirm title or lease control, zoning, permits, environmental conditions, access, easements, fiber routes, expansion rights, water supply, and community requirements.[9][12][13]
- **Power risk:** identify the executed agreements, megawatts and measurement point, delivery and ramp schedule, utility upgrades, deposits, energy supply, redundancy, curtailment, and remedies. Do not treat proximity or a queue request as delivered power.[9][10][17]
- **Construction and commissioning risk:** identify design responsibility, completion tests, schedule contingencies, long-lead equipment, contractor guarantees, insurance, and the consequences of late or partial delivery.[5][8][13]
- **Tenant and financing risk:** assess lease enforceability, counterparty credit, concentration, collateral, guarantors, lender conditions, and the project's ability to operate or relet after default.[8][14][15]
- **Technical and residual-value risk:** test whether power density, cooling, network access, and floor design can support future hardware or another tenant without costly conversion. Contractual residual support should not be confused with guaranteed market demand.[4][5][8]
- **Grid and public-impact risk:** examine upgrade cost allocation, reliability, electricity rates, emissions, water, noise, backup generation, and the effect of new tariffs or interconnection rules on the project and surrounding customers.[10][11][12][16]

## Terminology limits

LPS is best read as a scope prompt: which land rights, which power rights, and which shell obligations have actually been delivered? It is not a certification. The following distinctions prevent common overstatements:[5][6][9]

- **Capacity is not consumption.** Reserved or planned megawatts describe a right or design limit, while energy use depends on equipment, utilization, cooling, and operations.[11][16]
- **Contracted is not energized.** A power agreement can precede network construction, commissioning, and tenant acceptance by years.[9][10][17]
- **Shell is not turnkey.** A powered shell may exclude major cooling and electrical systems as well as all IT equipment.[5][6][13]
- **Credit support is not construction funding.** A guarantee may activate only after specified conditions and defaults, and its maximum exposure is not necessarily cash paid.[3][4][15]
- **Residual value is not assured resale value.** It is a contractual allocation of defined downside under stated remedies and caps.[4]

Because these boundaries vary, reliable descriptions should state the development stage, delivery date, IT versus facility power, included systems, responsible parties, and contingent terms. Using LPS without those details can obscure the very risks the category is meant to organize.[4][5][9][13]

## References

1. Nvidia. *Securing the Infrastructure of Intelligence*. August 17, 2026. https://blogs.nvidia.com/blog/securing-the-infrastructure-of-intelligence/
2. Nvidia. *NVIDIA Omniverse DSX Blueprint for AI Factory Digital Twins*. Accessed August 19, 2026. https://build.nvidia.com/nvidia/omniverse-dsx-blueprint-for-ai-factories/blueprintcard
3. Nvidia Corporation. *Annual Report on Form 10-K for the fiscal year ended January 25, 2026*. Filed February 25, 2026. https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm
4. Nvidia Corporation. *Current Report on Form 8-K*. August 17, 2026. https://www.sec.gov/Archives/edgar/data/1045810/000104581026000069/nvda-20260817.htm
5. CBRE. *Data Center Solutions*. Accessed August 19, 2026. https://www.cbre.com/services/property-types/data-center
6. Owen Rogers. *Retail vs wholesale: finding the right colo pricing model*. Uptime Institute, December 3, 2025. https://journal.uptimeinstitute.com/retail-vs-wholesale-finding-the-right-colo-pricing-model/
7. PGIM Real Estate. *Global Data Centers*. February 2021. https://cdn.pficdn.com/cms/pgim-real-estate/sites/default/files/2021-01/Global%20Data%20Centers%20February%202021%20PGIM%20Real%20Estate.pdf
8. CBRE Investment Management. *Data Center Investment: Decoding Opportunities*. Data current as of March 31, 2026 unless otherwise noted; accessed August 19, 2026. https://www.cbreim.com/insights/articles/decoding-data-centers
9. Linklaters. *Powered Land: An emerging strategy - Part 1*. June 2, 2026. https://www.linklaters.com/en/insights/thought-leadership/powered-land/powered-land-an-emerging-strategy
10. Federal Energy Regulatory Commission. *Interconnection of Large Loads to the Interstate Transmission System, Docket No. RM26-4-000*. Updated July 22, 2026. https://www.ferc.gov/rm26-4
11. Jessica Granderson, Ian M. Hoffman, Billie Holecek, Eliot Crowe, Sarah Josephine Smith, and Natalie Mims Frick. *Integrating AI Data Centers with the Power Grid*. The Bridge, May 2026. https://doi.org/10.20357/B7X61S
12. Jay Dietrich. *Water is a local issue: site selection and facility design*. Uptime Institute, August 8, 2024. https://intelligence.uptimeinstitute.com/resource/water-local-issue-site-selection-and-facility-design
13. Texas Tech University System, Texas Tech University, and Fermi SPE, LLC. *Ground Lease Agreement, Exhibit 10.9*. May 14, 2025. https://www.sec.gov/Archives/edgar/data/2071778/000121390025085175/ea025233301ex10-9_fermi.htm
14. Fermi Equipment Finance Holdco II, LLC and lenders. *Loan and Security Agreement, Exhibit 10.1*. February 19, 2026. https://www.sec.gov/Archives/edgar/data/2071778/000121390026020399/ea027781001ex10-1_fermi.htm
15. Applied Digital Corporation. *Current Report on Form 8-K/A*. Filed April 1, 2026. https://www.sec.gov/Archives/edgar/data/1144879/000149315226014498/form8-ka.htm
16. Arman Shehabi, Sarah Josephine Smith, Alex Hubbard, Alexander Newkirk, Nuoa Lei, Md AbuBakar Siddik, Billie Holecek, Jonathan G. Koomey, Eric R. Masanet, and Dale A. Sartor. *2024 United States Data Center Energy Usage Report*. Lawrence Berkeley National Laboratory, December 19, 2024. https://doi.org/10.71468/P1WC7Q
17. Federal Energy Regulatory Commission. *FERC Launches Aggressive Targeted Action to Speed Large Load Integration*. June 18, 2026. https://www.ferc.gov/news-events/news/ferc-launches-aggressive-targeted-action-speed-large-load-integration