The Cloud Is a Place

Mara has twenty minutes before she leaves for work. The letter on her kitchen table has been there for three days. It concerns a repair to her flat, includes a paragraph she cannot make sense of, and ends with a deadline that is suddenly close.

She takes a photograph of the page. An AI tool helps her pull out the dates and turn her questions into a clear reply. She checks the answer against the letter, changes a sentence, and saves it. The repair still needs arranging. The landlord still has to respond. But the task has moved.

This is a modest kind of progress. It deserves defending.

Now leave the kitchen. Follow the request beyond the phone, through communications equipment and into the machines that process it. Somewhere, physical equipment performs calculations. That equipment draws power. Heat must go somewhere. Wires, cooling systems, buildings and people keep the service available. Depending on the system and its location, water may help carry away heat. More water may have been involved in producing its electricity or making its equipment.

Mara can see the answer. Most of the arrangements that made it possible are out of view.

That distance is the subject of this book. We have learned to call an enormous physical system the cloud. The name is convenient. It becomes dangerous only when we begin to think like it.

A cloud seems to belong everywhere and nowhere. A substation occupies land. A water pipe has a source. A construction crew works a shift. A household pays a bill in a place with a regulator, a budget and a history. The service may travel across borders in seconds. Its physical obligations do not disappear at the border of the screen.

We should follow them.

Useful progress deserves a good foundation

There is an easy version of this argument in which the reader is asked to feel bad about using a new tool. The story begins with a helpful answer and ends with a finger pointing at the person who asked for it. You wanted help with a letter. Look what you have done.

I find that approach politically weak and often intellectually careless. It confuses an ordinary user’s small decision with the decisions that shape an industry. It also leaves people with fewer resources carrying a special burden of restraint while those with money keep buying convenience.

The household that needs help understanding a form has a legitimate interest in better tools. So does the mechanic reading unfamiliar technical instructions, the small business sorting a difficult pile of records, or the local group making its meeting accessible to more people. Each use needs judgment. An unreliable answer can create work rather than save it. But useful capability is worth building, and access to it should reach beyond the people who already have assistance.

At the same time, a good purpose does not settle the terms of its infrastructure. A building serving a worthy use still needs a safe electrical connection. A company helping some customers still owes an explanation to its neighbors. The promise of future benefit cannot decide who pays an unpaid construction bill today.

This is where a grown-up argument begins. We can want the tool and inspect the agreement. We can support the industry and refuse a careless project. We can ask for faster useful progress while demanding that the systems underneath it become more reliable, more accountable and easier for ordinary people to understand.

The ambition should be large. A society that becomes better at thinking, designing, discovering and solving problems ought to become better at meeting its physical needs too. We should expect stronger power systems, cleaner production, well-trained workers and public services that can afford to use what has been invented. An impressive answer on a screen is a poor finish line for that ambition.

The test reaches all the way to the kitchen table.

A large number needs an address

The International Energy Agency’s April 2026 outlook estimated that all data centers used 485 terawatt-hours of electricity worldwide in 2025. Its central projection put the 2030 figure near 950 terawatt-hours, about 3 percent of global electricity use. These are estimates and projections, not a census of every meter. They cover data centers as a whole; they are not an AI-only account.¹

You do not need to memorize those figures. Notice what has to travel with them: a place, a year, a unit, a boundary and a statement about how the number was obtained. Remove those details and a large number can be made to serve almost any argument.

Someone can point to a small global share and declare the issue unimportant. Someone else can point to a huge total and declare the entire enterprise intolerable. Both may be looking past the actual decision. A global percentage does not tell a town whether its water system can serve a new facility during a dry summer. An annual total does not tell a grid operator what can be delivered on a difficult evening.

The Republic of Ireland shows why concentration matters. Its official release covering 2024 recorded data centers using 6,969 gigawatt-hours, or 22 percent of total metered electricity. That is a country-specific, all-data-center statistic from that release, not a global AI share or a measure of all energy use.²

The same kind of activity can be a modest share of one system and a major part of another. There is no contradiction. The systems differ.

Imagine two towns considering an identical new building. One has an appropriate site, room in the relevant network, a workable water plan and a customer prepared to guarantee its obligations. The other has a strained connection, homes close to the proposed equipment and no settled answer about who funds the upgrade. A national average cannot approve both projects or reject both projects. It cannot see the difference that matters.

The address is part of the answer.

This is also why comparisons with households need care. A headline may say that a facility uses enough power for a city. Which city’s households? In which climate? Does the comparison concern a peak rate or a year’s energy? Is it based on the building’s maximum connection or its actual use? A dramatic comparison can help us picture scale. It can also quietly turn an uncertain plan into an operating fact.

We owe the reader something sturdier than an image that wins an argument by hiding its assumptions.

The people on both sides of the fence

When a proposed development enters a public meeting, people are often sorted into two groups before they have said much. There are the people who welcome jobs and the people who oppose the future. Or there are the people who defend a community and the people who sold it.

These are convenient characters. Actual people are harder to arrange.

A construction worker may want the project and worry about its effect on the family power bill. A retired teacher may oppose the current site and support the technology. A renter may receive little direct benefit from a property-tax change while relying on services funded by local revenue. A small business owner may need better digital tools and dependable, affordable electricity. One household can contain all of these interests.

Treating them as a single vote for or against technology wastes useful knowledge. The worker knows whether training promises lead to paid work. The neighbor knows what the road feels like at school pickup. The water operator knows where a reassuring annual average stops being useful. The grid engineer knows which connection cannot be made safe by a speech about urgency.

None of these people knows everything. Neither does the company. Good public decisions bring their knowledge into contact before a bad assumption hardens into concrete.

The fence around a facility is a property boundary. It is rarely the full boundary of its consequences. Some benefits leave it as services, wages and taxes. Some costs leave it as demands on shared systems. Other effects depend on the design: where equipment is placed, how it is cooled, when it runs and what happens if a plan changes.

We need to be precise about those paths. If every concern is thrown into one bucket called the environmental cost of AI, we lose the ability to solve distinct problems. A charge designed to fund electrical equipment will not quiet a poorly placed machine. A noise rule will not protect a utility against a customer’s departure. A training program will not establish a sustainable water supply.

Different obligations need different tools. That is less dramatic than a single verdict. It is much more useful.

Four accounts on one kitchen table

Return to Mara. Suppose she now hears that a large data center may be built outside town. She has four separate questions, though nobody has explained them that way.

What will the operator pay for its own electricity and equipment?

What might change in the shared system that serves her home?

What will the town receive, spend or give up?

What will happen to the people nearest the site?

A payment in one account does not automatically settle another. A company may pay its electricity supplier in full while a dispute remains about a network upgrade. A town may collect new taxes while one neighborhood bears a burden the budget does not measure. A developer may fund a local program while leaving an unclear obligation if the project is abandoned.

These are possible arrangements to investigate, not accusations against every development. In a well-designed agreement, many of the questions can be answered well. That is the point of asking them early.

A useful public discussion should let Mara follow each account without becoming a specialist in all of them. She should be able to find a plain explanation of the load plan, who funds required works, what the community has been promised and which office can enforce the promise. Technical details should remain available for people qualified to test them.

Plain language and technical rigor belong together. The public summary helps people enter the argument. The underlying record keeps the summary honest.

When a proposal cannot survive that combination, the problem is larger than public relations.

The bill after the excitement

Construction has a visible beginning. Long obligations are quieter.

A new connection may require equipment built for years of service. A contract can outlast the executive who signed it. An elected official may leave office before the promised tax revenue reaches its expected level. A customer can change plans because its technology, financing or market changes. None of that automatically means somebody acted badly. It means time is part of the agreement.

Households understand this. A person taking on a long commitment asks what happens if income falls, an appliance fails or a family member needs help. An institution should be at least as serious when it commits shared resources.

The weak bargain assumes that everything will go as advertised. The stronger bargain decides what happens when it does not.

Who pays if demand arrives late? Who pays if it is smaller? What happens if the company is sold? Which obligations remain attached to the site? Can the town afford to keep the services it began funding? If a cooling plan must change during a drought, who has the authority to require it?

These questions need not be hostile. A responsible company also benefits from clear terms. Uncertainty about the public agreement can become uncertainty about the business. A project that has dealt honestly with its neighbors and its obligations has a better foundation than one relying on everyone staying too busy to look closely.

We should make responsibility easier to demonstrate and harder to fake.

A different meaning of being pro-progress

Supporting progress should mean more than defending whatever is already being proposed. It should involve a view of what better would look like.

Better might mean useful computation arriving alongside new dependable supply rather than competing indefinitely for the same narrow opening. It might mean an apprentice gaining a qualification that remains valuable after the building is finished. It might mean a town using uncertain new revenue to strengthen its finances rather than making promises it cannot keep. It might mean a household getting a tool that saves time without quietly becoming the backup payer for a much larger customer’s gamble.

These outcomes will not appear because we use the word innovation often enough. They require decisions about ownership, contracts, public capacity and technical design. Some decisions belong to a utility regulator. Others belong to a water supplier, a planning authority, a legislature or the company itself. A serious argument identifies the actor instead of asking the atmosphere to improve.

There will be trade-offs. Some good projects will cost more when their full obligations are recognized. Some designs will need to move. Some timelines will be slower than a developer wants. Other projects may move faster because their commitments are clear and their location fits the system. We should judge the bargain by what it delivers, not by whether every participant gets an effortless yes.

We should also be willing to build. Avoiding every new demand can become its own form of neglect. A system with too little capacity can ration opportunity, preserve old inefficiencies and leave public needs unmet. The answer to careless building cannot be permanent helplessness. It must be the skill and public authority to build well.

That is the stance of this book. I want more useful capability, and I want the obligations beneath it to be visible. I want the people who make the future work to have a say in its terms. I want promises that can survive a change of owner, a disappointing quarter and a quiet room after the ceremony is over.

Mara’s answer took a moment. The institutions that make such moments possible require longer attention.

The cloud is a place. Several places, often. People live there, work there and pay bills there. We can follow the connection until they come into view.

Then we can decide what we owe one another.