What the Connection Really Promises

At the meeting, the number sounds decisive.

The proposed campus will need one hundred megawatts. Someone writes it on a slide. A resident photographs it. By the next morning, the number has become the whole project. People repeat it as though it tells them what will be built, how much electricity will be used and what their bills will become.

It tells them none of those things by itself.

One hundred megawatts could describe a requested connection, an agreed ceiling, the power drawn during a particular interval or a future stage that has not yet been financed. Those are related quantities. They are not interchangeable. The difference is where a great deal of public risk can hide.

Before we argue about whether the number is large, we need to ask what the number is doing.

The rate and the amount

Power is the rate at which energy is being used or delivered. Energy is the amount over a period. The distinction is familiar even if the units are not.

A tap can run fast for a short time or slowly for a long time. Its flow rate and the amount collected answer different questions. Electricity has the same useful distinction, though an electrical system has constraints that a bucket does not capture.

A watt measures power. A kilowatt is a thousand watts. A megawatt is a thousand kilowatts. When a one-kilowatt device runs steadily for one hour, it uses one kilowatt-hour of energy. The hour is doing important work in that sentence.

Now take a deliberately simple example. A facility draws ten megawatts steadily for two hours. It uses twenty megawatt-hours. A different facility draws twenty megawatts for one hour. It also uses twenty megawatt-hours. Their energy totals match. Their demands on a connection at a moment do not.

If we need equipment that can carry the higher rate, the same energy total does not erase that requirement. If we care about energy use across a month, the peak alone does not answer the question either. We need both, plus information about timing.

This is why a gigawatt cannot be added to a terawatt-hour. One is a rate. The other is an amount. It is also why a facility’s maximum possible demand should not be treated as though it ran at that level every hour. You can calculate that hypothetical case, but you have to call it hypothetical.

The units are a form of public protection. They make a claim harder to stretch without being noticed.

The space held open

Capacity introduces another question. How much service has the system agreed to keep available?

Suppose a new customer wants room to grow. It asks for a connection larger than its opening-day needs. That can be a reasonable business plan. Installing everything twice may be wasteful. An operator may also need spare capacity for reliability or later equipment.

But keeping room available can have a cost before the customer uses it. Equipment may have to be sized, land obtained, studies completed or alternative connections reconsidered. The promise to serve is itself a commitment, even while the meter is quiet.

Think of a workshop booking a delivery vehicle for a full day. The vehicle cannot necessarily be promised to another customer just because the first workshop’s shipment is late. The service held ready has value and cost. The analogy ends before the engineering begins, but it reveals the commercial issue: actual use and reserved service can create different obligations.

A household usually has little control over how a very large customer negotiates those obligations. That is why public rules matter. If the customer receives an option to use expensive capacity later, who is paying to keep that option open? If the customer walks away, who owns the equipment and who remains responsible for it?

The correct answer may differ across projects. Some equipment can readily serve others. Some works would have been needed anyway. Some investments are useful to the wider system. Other works are closely tied to the particular request. A fair account must make those distinctions rather than presenting every cost as either entirely private or entirely public.

We should not punish a customer for sensible planning. We should price and govern the planning commitment honestly.

Four questions about one connection

Scroll sideways to see all columns. Keyboard: focus the table and use the left and right arrow keys.

Text equivalent: One connection, four questions

Part of the figure	Meaning
Book	The Cloud Has a Power Bill
Title	One connection, four questions
Starting point	Name the question before comparing the number.
Question group	Four different accounts
Reserved capacity	What amount has the connection agreement set aside? Unit: megawatts (MW)
Measured power	What rate did the meter record at the stated time or interval? Unit: megawatts (MW)
Energy over time	How much electricity was used across the chosen period? Unit: megawatt-hours (MWh)
Dependable supply	What can the system reliably deliver in a stressed hour? Unit: megawatts (MW)
Reminder	Before comparing numbers, name the boundary, time period and agreement.
Status	Original conceptual map. No project measurements or forecast.
How to read the diagram	The four panels ask different questions. They are not stages, alternatives in a forecast, measured amounts or quantities to add together. The three power or capacity questions share a unit, but they are not interchangeable.

Figure 1. One connection, four questions. An original conceptual map of reserved capacity, measured power, energy over time and dependable supply in a stressed hour. Equal panels do not imply equal quantities. This explanatory aid is not measured project data or a validated planning model.

The diagram separates questions that public discussion often squeezes together. What has been reserved? What is being drawn? How much has been used over time? What can be supplied when conditions are difficult?

The last question deserves special attention. A system can produce plenty of electricity across a year and still face a hard hour. Equipment can be unavailable. Weather can change demand and output. A connection can be constrained even while supply exists somewhere else. We need the ability to deliver appropriate service at the time and place promised.

That does not require every customer to solve the entire system alone. Networks exist partly because sharing can be useful. It does require clarity about the service being sold. A connection that can be interrupted under agreed conditions is different from one promised continuous firm service. A price should reflect the actual obligation, and the customer should understand what interruption means for its work.

A facility may have batteries or backup equipment. Their presence is relevant, but the label backup does not tell us how much useful support they can provide to the wider system. How long can they operate? Under what conditions? Can they serve only the facility, or export electricity? What permits and technical rules apply? What fuel or charging assumptions are being made?

An impressive equipment list is the beginning of an inquiry. It cannot replace the operating plan.

The proposal has stages

Imagine the campus developer says it could eventually need one hundred megawatts. The word eventually covers several different decisions.

First comes interest in a location. Then a request for information or a study. Land may be controlled, financing pursued and designs changed. An agreement may establish what will be delivered, when and under which conditions. Construction follows its own sequence. Equipment is installed. The first portion of the site begins operating. Later portions may arrive gradually.

No universal sequence fits every jurisdiction or company. The important point is that an early request and a running facility occupy different stages of commitment.

A town should ask what evidence belongs to each stage. A signed service agreement means more than a conversation. A financial guarantee means something different from an optimistic forecast. A meter reading means something different from an advertised maximum. None should be dismissed merely because it is incomplete. Each should be given the weight it has earned.

This is especially important when the same business considers several locations. Looking at alternatives is normal. Counting every alternative as though all will be built is a planning error. The task is to distinguish competition among possible sites from separate committed demand, while protecting genuinely confidential commercial information.

Public planners need a way to ask the question without assuming fraud. Developers need a way to provide credible evidence without disclosing every part of their business. A well-designed process can make both possible.

The public needs the result: an account of what is firm, what is conditional and what would change the plan.

Growth is a path, not a switch

Opening a building does not necessarily bring it immediately to its full planned demand. A load ramp describes how use is expected to increase over time. It might rise in steps as new sections become ready. It might be slower or faster than expected.

For our imagined campus, suppose the first stage is small and later stages depend on customers that have not yet signed. The town should not treat every stage as equally certain. It should also avoid the opposite mistake of assuming only the first stage matters. The first agreement may commit resources needed for the later stages.

The useful question is what happens at each step. Which works must be completed before the first connection? Which can wait for stronger evidence? What payment is due while capacity is being held? When must an unused reservation be released? If the schedule slips, who is told and what changes?

These questions can improve the project. A staged plan may avoid spending too early. It may also expose a false economy: delaying one component could make later expansion much more costly. That is an engineering and financial question to examine, not a reason to pretend the future is certain.

There is a difference between leaving room for growth and making everybody else underwrite every possible version of it.

The distinction should be visible in the agreement.

What the national model can tell us

Berkeley Lab’s 2025 Update, published in June 2026, estimated U.S. data-center electricity use at 192 terawatt-hours in 2024. Its reference case reached 649 terawatt-hours in 2030; combined low and high assumptions produced 521 to 843. These are bottom-up model results for U.S. data centers, excluding cryptocurrency. The endpoints are scenario bounds, not probabilities or an AI-only forecast.¹

A model like this helps us think about scale and uncertainty. It does not issue a building permit. It cannot tell our imagined town which substation will be ready or whether a particular customer’s promise is secure.

Nor should we splice the U.S. figures into the global outlook from the previous chapter as if they were one measuring instrument. Different studies have different boundaries and assumptions. A careful reader can learn from both without forcing them into false agreement.

This is a useful habit beyond data centers. When two forecasts differ, first ask whether they are forecasting the same thing. Only then ask why the answers differ. Otherwise, we can spend a long time debating a contradiction we manufactured ourselves.

The local plan should be informed by broader evidence, but it still needs local proof.

A readable connection file

What would that proof look like for a resident who has a job and no spare week to study utility filings?

I would want a short public record with a few clear distinctions. The customer has requested this amount. The current agreement reserves this amount. The first operating stage is expected at this date. These investments are committed now. These later investments depend on specified milestones. This office reviews changes. These parts of the record are confidential for stated reasons.

Behind the summary would sit the engineering and commercial detail. Independent reviewers would need access appropriate to their job. Residents should not have to choose between a cheerful leaflet and an unreadable archive.

The file should also show what the agreement does not establish. A power connection does not settle water availability. A land-use approval does not prove a future revenue estimate. A plan for backup generation does not automatically authorize unlimited operation. Each decision has a boundary.

This is not a demand for endless documents. It is a demand for a usable chain of responsibility. If nobody can explain which decision answers which question, the project may have more unresolved issues than its presentation suggests.

The most important sentence in a public record is sometimes the one that admits what remains conditional.

The service already promised

A connection discussion can become so focused on the new arrival that existing service disappears from view. The utility is not starting with an empty map. It has people who already depend on it and commitments that must remain credible.

Imagine a workshop planning to add equipment next year. Its owner has heard that the new campus has secured a large reservation. She wants to know whether her own smaller expansion will still be possible. The answer should come from a transparent process for allocating and planning service, rather than whoever can make the most impressive announcement.

Different requests may justify different treatment because their readiness, technical effects and service needs differ. But the process should make those distinctions understandable. An early speculative request should not automatically receive the authority of an operating commitment. A small customer should not become invisible simply because its individual demand is less newsworthy.

The same principle applies to maintaining existing infrastructure. Growth should not distract from the work that keeps current service dependable. A new facility can contribute to a stronger system; its arrival should be examined alongside the needs already there.

For public review, I would ask for an account of how the proposal fits the wider service plan. Which other needs were considered? Which alternatives were tested? What remains available for credible future uses? What would cause the plan to be revised?

These questions do not demand a reservation for every imagined future need. They ask the planner to show that one visible customer has not become the whole picture. The people already connected should remain present in the decision about the next connection.

Leave room without leaving a blank check

There is a legitimate objection to all this caution. New industries cannot be built only after every future customer is known. Infrastructure requires judgment ahead of demand. If we insist on perfect certainty, we will get neither reliable expansion nor useful innovation.

That objection is right as far as it goes. Certainty is unavailable. The answer is to decide how much uncertainty to accept, what evidence justifies it and who carries the consequences.

A community may choose to share a measured risk because it expects a genuine public benefit. A utility may make a prudent investment whose value extends beyond one customer. A developer may accept firm payment obligations in return for dependable access. These can be good bargains when their terms are visible and their downside is survivable.

The trouble begins when one participant receives the upside of flexibility while another inherits the cost of disappointment without having agreed to it.

A household asked to support growth deserves to know the form of that support. Is it funding an asset everyone needs? Taking a temporary risk for a defined public return? Or absorbing a private option that has been described as inevitable progress?

These are political questions expressed through technical arrangements. Learning the units will not answer them all. It will let more people ask them properly.

The meeting’s one hundred megawatts can now become a series of useful questions. What stage? What commitment? What use? What service under stress? What payment if the plan changes?

The number has become less magical and more informative.

That is progress too.