AI compute hosting · Guide 17 of 17

Plain-language definitions for AI nodes, offtake, powered land, behind-the-meter power and more.

80%host share of compute and storage revenue
$9,689est. monthly net, one RTX PRO 6000 server
9.4 kWsite power per RTX PRO server

*Estimates using current program numbers (730 contracted hours a month). Not an offer or a promise of income.

Updated November 18, 2026 by Wolf River Electric.

Short answer: AI and energy now share a vocabulary. An AI node is a GPU server that rents its computing power to AI customers. Offtake is who buys the power a solar array makes. Behind the meter means the generation and the load are on your side of the utility meter. Powered land is a site with power already in place. SI is short for superintelligence, the goal some AI labs now talk about openly. Here are the terms you will hear, in plain language. How AI compute hosting works.

AI and compute terms

AI compute

The processing power used to train and run AI models, mostly supplied by GPUs in data centers. AI compute hosting.

AI node

A single AI server, or a small group of them, connected to a cloud network that rents its GPU time to companies. In our programs, an AI node is typically an 8-GPU server cabinet on your property.

AI site hosting (host)

Providing the space, power, cooling and internet for AI nodes and earning 80% of what they make under current terms. The site owner who owns the servers is the host.

AI factory

A term popularized by Nvidia for data centers built to produce AI at scale, where the output is often measured in tokens.

Edge inference

Running trained AI models close to where they are used, instead of in a distant central data center. Distributed AI nodes are one form of the edge.

GPU

Graphics processing unit, the chip that does most AI work. Server GPUs such as the Nvidia RTX PRO 6000, B200 and B300 are what AI nodes rent out.

GPU-as-a-service

Renting GPU time by the hour instead of buying hardware. AI nodes earn from this market.

Inference vs training

Training builds an AI model; inference is using it to answer questions or make images. Much of the growth in AI demand is inference.

Neocloud

A newer cloud provider that specializes in renting GPUs for AI, as opposed to the large general-purpose clouds.

SI (superintelligence)

A hypothetical AI that would exceed human ability across most tasks. Some AI labs describe it as their goal, which is part of why they want so much more compute.

Sovereign AI

AI infrastructure and models kept under a country's or organization's own control and borders.

Tokens

The small pieces of text an AI model reads and writes. AI usage and AI factory output are often measured in tokens.

Utilization

The share of hours an AI node is actually rented. It is the biggest driver of hosting revenue. ROI examples.

Distributed or decentralized compute

Computing spread across many independent sites instead of one large data center. DePIN, short for decentralized physical infrastructure networks, is a related term for networks built from infrastructure supplied by many owners.

Ground-mount solar tables in autumn grass, installed by Wolf River Electric
Generation on your own land is the energy side of distributed AI compute.

AI infrastructure supercycle

A long wave of spending on data centers, chips, power plants and grid upgrades to meet AI demand. Big Tech alone was projected to spend about $725 billion on AI infrastructure in 2026 (Techstrong.ai, May 2026).

Hyperscaler

One of the giant cloud companies, such as Amazon, Microsoft, Google and Meta, that build data centers at massive scale.

Capex

Capital expenditure: money spent on long-lived assets like servers, buildings and power equipment.

Picks and shovels

Investing in the tools a boom depends on rather than the boom itself. In AI, power and infrastructure are the picks and shovels.

Watts to tokens

Shorthand for turning electricity into AI output. Every token an AI model produces takes power, which is why energy sites matter to AI.

Energy and power terms

Behind the meter (BTM)

Generation, storage and loads on the customer side of the utility meter. Solar that powers your own AI nodes is behind-the-meter use.

Speed to power

How quickly a project can get the electricity it needs. With utility connections for large loads taking years, it has become a top priority for AI builders.

Powered land

A site with power capacity already secured. A powered shell is a building with power and basic infrastructure ready for a tenant's equipment. AI-ready solar farms.

Offtake and offtake agreement

Offtake is who buys a generator's output. An offtake agreement is the contract that sets the buyer, price and term. Energy offtake explained.

Power purchase agreement (PPA)

The most common offtake agreement, where a buyer purchases a solar array's output at a set price for years. A virtual PPA is a financial version tied to power delivered to the grid.

Net metering

A utility arrangement that credits exported solar against power you use. Rules and credit rates depend on system size and utility.

Avoided cost

What it would cost a utility to produce or buy the power itself. Larger solar systems are often paid a rate based on it.

Interconnection queue

The line of projects waiting for a utility to study and approve a grid connection.

kW vs kWh

A kW is power at a moment; a kWh is energy over time. An AI server drawing 7 kW for an hour uses 7 kWh.

Microgrid

A local set of generation, storage and loads that can run connected to the grid or on its own during an outage.

Energy park

A site that groups generation, storage and the loads that use the power, sometimes including data centers.

UPS

Uninterruptible power supply, a battery unit that keeps servers running through the first moments of an outage. Battery storage for AI compute.

Rack-mounted 8-GPU AI compute server with front drive bays and hot-swap power supplies
An AI node: one rack-mounted GPU server, like this 8-GPU unit.

Business and compliance terms

Passive income

Income that takes little day-to-day work once set up. AI hosting can be low-effort, but it is not guaranteed and depends on demand. What AI hosting really pays.

Revenue share

The split of AI compute revenue. Under current program terms the owner of the server, who is also the host, keeps 80%, and the integrator and systems administrator that installs, runs and maintains the servers keeps 20%.

SOC 2

An independent audit of how a service organization protects customer data. Cloud operators carry it, and host sites support it with physical security. Site requirements and SOC 2.

Infrastructure build

The electrical service, circuits, backup power, cooling connections, solar and batteries a site needs before AI nodes can run. That is the part Wolf River Electric builds.

Put the terms to work

We can show you what solar generation and AI nodes would look like on your property. Request an AI compute site assessment or call (763) 229-6662.

AI compute packages

One 8-GPU server each. Estimated using current program numbers.

Before loan payments and taxes, at 730 contracted hours a month and 10¢ power. Rental demand and rates can change; revenue is not guaranteed. Compare packages

Next guide: Overview →

Free quote

See what your site could earn hosting AI compute

Get a free AI compute site assessment from Wolf River Electric: power, solar, hosting space and fiber. No cost, no obligation.

Get my free quoteCall (763) 229-6662