On 2 October, Amazon announced Built Together, promising more than $1 billion over five years in additional support for US communities hosting its data centres. Education, job training, energy affordability and water preservation are among its priorities. This is a commitment, not evidence of benefits already delivered. But it exposes a question usually missing from discussions about artificial intelligence: what does society owe the people living beside the machinery?

AI can be useful to someone thousands of kilometres from the electricity supply and cooling water that make it possible. That distance separates beneficiaries from those negotiating the physical consequences. My contention is simple: communities need enforceable agreements about resources, costs and opportunity before they become hosts. A generous programme can strengthen that bargain; its headline value cannot settle it.

Consider a household that never pays for an AI subscription but shares a utility with a vast computing facility. Its interest is reliable, affordable electricity. A local electrician may welcome the same project because it offers skilled work. Both interests are legitimate. Describing the community as either enthusiastic or hostile conceals who gains, who pays and when.

The strongest case for building is substantial. Computing infrastructure supports useful services; construction can create work, and tax revenues can finance public needs. Virginia’s legislative audit commission, in its 2024 study, found significant economic benefits, concentrated particularly in construction. It also found that ongoing data-centre jobs can pay well. Rejecting projects indiscriminately can mean rejecting real opportunities.

Yet a construction boom and a durable employment base are different promises. Training should leave residents with credentials useful to several employers, including after a building is complete. Investors and governments should ask separately about temporary work, permanent employment and who can actually qualify. A scholarship offers little immediate protection to a pensioner facing higher living costs.

Electricity creates another distinction. The Virginia audit found that existing rates allocated current costs appropriately, while warning that expanding demand could still increase system costs and leave customers exposed if infrastructure was built for demand that failed to arrive. Paying today’s bill does not resolve tomorrow’s risk. That is a lesson about contract design, not proof that every data centre raises household prices.

The International Energy Agency’s 2025 base case projected global data-centre electricity consumption at around 945 terawatt-hours in 2030. That includes services beyond AI and remains a forecast. For a town, however, the decisive number is the capacity available locally during a difficult week. Global percentages cannot answer that question.

India should treat this as a planning opportunity. A February study by CEEW and Systemiq found explicit sustainability provisions in only five of the 15 state policy frameworks it reviewed. That is a dated research finding, not a claim about every policy in force today. It nevertheless shows why competition for investment needs to include resource planning alongside incentives.

For an Indian city, success should mean more than attracting a large capital commitment. It should mean power and water arrangements that work through seasonal stress, accessible information in local languages, and skills programmes residents can realistically attend. Hosting servers does not by itself ensure affordable computing for a small enterprise nearby. Access and local employment require deliberate choices.

Three futures follow from different bargains.

In a constructive future, operators help finance additional infrastructure, residents can inspect performance, and portable skills turn a building project into wider opportunity. Predictable obligations can also benefit investors by reducing conflict and delay.

In an uneven future, selected households receive training or efficiency improvements while others share constrained resources. The announcement succeeds on its own terms, yet people outside the programme remain worse placed.

In a harmful future, governments compete by weakening conditions, resource pressure mounts, and public utilities inherit costs when promised demand disappoints. This is a risk to guard against, not a prediction about Amazon’s programme.

The choice is practical. Publish site-level resource plans. Identify who pays for new capacity and who bears the loss if it goes unused. Give residents a route to challenge failures. Assess community funding alongside these obligations, with separate accounts for each. People should be able to welcome investment without having to accept an undefined liability.

Would the project still earn an acceptable return if it paid its full local costs? Which promised benefits survive the construction phase? Can neighbours verify the commitments and obtain a remedy when they are missed? Those answers belong in the investment decision, while the community still has a meaningful choice.