The Race to Control AI Begins With the Data Centre

Who Controls the Intelligence? · Part 1

Originally published: July 19, 2026
Fully updated: July 28, 2026
Status: Infrastructure and public-policy analysis

The global race for artificial intelligence is also a race for electricity, chips, cooling, land, fibre, construction capacity and control over who receives advanced computing power.

The artificial-intelligence race is usually described through models, benchmarks and promises of systems more capable than anything that came before them. But no model operates in the air. Every answer, image, video, song, analysis and automated task depends on physical infrastructure: chips inside buildings connected to electricity, cooling, communications networks and capital.

That changes the central question. The issue is not only who develops the strongest model. It is who can secure enough computing capacity to train it, operate it, improve it, protect it, price it and decide who may use it.

Control of that capacity can shape which researchers gain access, which businesses can compete, which governments remain dependent on foreign providers and which communities are asked to supply the land, power, water and public support behind the system.

This article does not argue that every data centre harms its community, that every public investment is unjustified or that one company already controls advanced intelligence. It argues that AI’s physical foundations now deserve the same scrutiny as the models built on top of them.

The community conversation that started this investigation

I began asking these questions after hearing Erin Brockovich speak with Theo Von about data centres and the concerns being raised by nearby communities. Residents were describing pressure on electricity systems, water use, noise and decisions moving forward before the people living closest to the projects understood what had been promised.

The conversation was useful because it translated a technical infrastructure story into a citizen question: what is being built, what will it require and who has agreed to provide it?

Brockovich’s public reporting initiative distinguishes among proposed, under-construction, operating and community-reported facilities. That distinction matters. A community report is evidence that a concern has been submitted. It is not automatic proof that a specific facility caused the alleged harm.

That is the standard this series follows: concerns deserve investigation; claims still require verification.

Supporting AI does not require surrendering public scrutiny

I use artificial intelligence. I teach creators how to work with it. I have experienced what it can make possible for people who lacked money, access, technical training or confidence to begin creating.

AI infrastructure can also support medical research, accessibility, cybersecurity, scientific modelling, public administration and domestic technological capacity. Those benefits are real enough to take seriously.

They are not a reason to accept every subsidy, utility agreement, environmental claim, planning decision or corporate projection without examination.

Supporting AI does not require blind support for every company, investor, executive, utility agreement or government decision attached to it.

This is not a question owned by the political left or right. Electricity reliability, property rights, water security, tax incentives, domestic capacity, national dependence and transparent government contracts can matter across political traditions.

The cloud is a building

The word cloud makes digital services sound weightless. They are not.

When a person sends a request to an AI system, a physical chain begins:

1. The request travels through communications networks.
2. Servers and accelerators process the model’s calculations.
3. Storage and networking systems move data between machines.
4. The result travels back to the user.
5. Electricity, cooling, security and operations sustain every stage.

Inside and around the facility are accelerators, servers, storage systems, networking equipment, fibre connections, transformers, substations, batteries, uninterruptible power systems, backup generation, cooling equipment, monitoring systems and security controls.

Some facilities use evaporative cooling and substantial quantities of water. Others rely more heavily on air, closed loops, reclaimed water or hybrid systems. A statement about “data-centre water use” is incomplete unless it identifies the cooling design, local climate, operating load and whether the figure describes withdrawal, consumption, permitted maximum or expected average use.

The global percentage can hide the local problem

The International Energy Agency estimates that data centres consumed about 415 terawatt-hours of electricity in 2024, approximately 1.5% of global electricity consumption. Its base case projects demand of roughly 945 terawatt-hours by 2030—just under 3% of global demand.

Those numbers do not mean that all data-centre electricity is used for AI. They do show the scale of the wider infrastructure system within which AI demand is expanding.

The IEA expects accelerated servers, driven primarily by AI workloads, to account for almost half of the net increase in global data-centre electricity use through 2030.

The global percentage can still mislead. Data centres do not spread evenly across the planet. They cluster around particular transmission regions, substations, fibre routes, tax environments and communities. A modest global share can create an extraordinary local request when several large campuses seek power from the same grid.

Training a model and operating it are different infrastructure demands

Training

Training a frontier model can require a large, tightly connected cluster of advanced chips operating for weeks or months. It is a concentrated effort that produces or substantially updates a model.

Inference

Inference is the continuing computation required whenever people, businesses or automated systems use that model. Demand can rise with more users, longer outputs, image and video generation, real-time services and agents that complete many steps instead of answering one question.

A company may complete one major training run, but the electricity and computing demand continues every time millions of people use the resulting system.

That is why the race does not end when a laboratory announces a breakthrough. The model must still be served, maintained, secured and improved.

Why construction is accelerating

Public demand is part of the answer. Generative AI is being integrated into software, customer service, coding, research, design, logistics, media production and decision-making. New agent systems may remain active longer and consume more computation than a single chatbot exchange.

But consumer demand alone does not explain the urgency. Governments and companies increasingly describe computing capacity as strategic infrastructure tied to economic power, national security, scientific competitiveness and technological sovereignty.

That framing changes what governments are willing to provide. The request is no longer simply for a private building permit. It can include faster grid access, new transmission, tax concessions, public procurement commitments, planning support and national programs designed to secure domestic compute.

Stargate: from announcement to construction

In January 2025, OpenAI and SoftBank announced Stargate, a company intended to invest up to US$500 billion over four years in American AI infrastructure, beginning with an intended US$100 billion deployment.

The announcement connected the project to American AI leadership, national security and the development of artificial general intelligence. It also made clear that power, land, construction and equipment were not side issues. They were the project.

By 2026, the story had moved beyond a single announcement. OpenAI had identified additional Stargate sites and partnerships, established an initial target of 10 gigawatts by 2029 and begun construction on a one-gigawatt campus in Saline, Michigan, known as “The Barn.”

That progress should not be confused with completion. The full US$500-billion figure remains a future investment target, not money already spent. A planned gigawatt is not the same as energized computing capacity.

OpenAI has said the Michigan project will be structured so local residents do not bear its infrastructure and energy costs. That is a company commitment worth tracking against final utility agreements, rate treatment and public reporting after the campus opens.

The international race for sovereign compute

The United States is not alone. Canada, the European Union, the United Kingdom, India and other jurisdictions are building programs intended to reduce dependence on foreign computing capacity.

These programs do not share one ownership model or political purpose. Some rely on public supercomputers. Some subsidize commercial capacity. Some coordinate planning and power access. Some are designed to broaden access for researchers and smaller firms.

Canada’s sovereignty question is bigger than geography

Canada’s 2026 national AI strategy treats compute, cloud and connectivity as foundations of sovereignty. It projects a large increase in commercial demand by 2030 and acknowledges that much of the capacity available to Canadian organizations is currently foreign-owned or foreign-controlled.

The federal strategy includes public supercomputing, commercial domestic capacity and an access fund intended to help researchers, startups and smaller firms obtain compute they could not otherwise afford.

For Canadians, the important question is not only how many megawatts are built inside the country. It is what “Canadian” control means after the ribbon is cut.

• Who owns the land and building?
• Who owns or leases the chips?
• Which cloud platform controls the software layer?
• Who decides which customers receive priority?
• Where does sensitive data fall under law and contract?
• Can Canadian researchers retain access during geopolitical or commercial conflict?
• What public return is enforceable when public money supports the project?

A data centre located in Canada is not automatically sovereign if its cloud platform, hardware, financing, operating decisions and customer priorities remain controlled elsewhere.

What sovereign compute actually means

The term is often used as though it has one definition. It can describe several different kinds of control:

• Physical sovereignty: the hardware is located inside the country.
• Legal sovereignty: operations fall under domestic law.
• Ownership sovereignty: domestic entities own the infrastructure.
• Operational sovereignty: domestic institutions can operate it without foreign permission.
• Data sovereignty: sensitive information remains under domestic protections.
• Access sovereignty: researchers and businesses cannot be cut off by a foreign provider.
• Supply-chain sovereignty: chips, maintenance and software are not wholly dependent on external sources.

A government can achieve some of these without achieving all of them. Responsible reporting should identify which form of sovereignty a project actually provides.

Creating a model is only the beginning

Public discussion often treats the race toward advanced AI as though it ends when a laboratory produces a breakthrough system. It does not.

Creating a powerful model is one achievement. Maintaining enough chips, electricity, cooling, networking, security, engineers and financing to serve millions of users is another.

Access can be interrupted by export controls, grid delays, transformer shortages, financing failures, cyberattacks, contract disputes, government action and international conflict.

This infrastructure race can continue even if the most ambitious predictions about artificial general intelligence prove wrong. The same facilities can support cloud services, scientific modelling, government systems, media production and less dramatic forms of AI.

Who actually owns a data centre’s power?

A data-centre campus is rarely controlled through one simple ownership relationship. The entities involved may include:

1. Landowner
2. Building developer
3. Facility operator
4. Utility customer
5. Cloud provider
6. Chip owner or lessee
7. Model developer
8. Financial investor
9. Government partner
10. End customer

“The data centre belongs to Company X” may hide several different owners, contracts and forms of control.

That matters because accountability depends on knowing which entity received the tax concession, signed the power agreement, obtained the water permit, controls the computing capacity and decides who gains access.

When computing capacity becomes power

Stanford’s 2026 AI Index reports that private industry produced more than 90% of the notable frontier models identified in 2025. The same report describes a hardware and infrastructure system concentrated around a relatively small number of leading providers.

That concentration matters because computing capacity can determine who is able to train frontier systems, test safety claims, conduct advanced research and compete with established companies.

A university may have talented researchers without enough compute to reproduce a corporate result. A startup may have a useful idea without the capital or cloud access required to test it at scale. A country may possess strong institutions while depending on foreign firms for the infrastructure underneath them.

Compute can therefore become corporate and national influence. It can affect which governments gain intelligence or military capability, which businesses receive affordable access, which researchers can independently evaluate powerful models and which institutions set the acceptable conditions of use.

This is a supported inference—not proof that one company already controls advanced intelligence.

The Manhattan Project comparison helps—and then breaks

The comparison is useful in three ways: strategic competition, government mobilization and scarce technical expertise tied to major physical construction.

It breaks when treated as a description of ownership. The Manhattan Project had a defined wartime objective and a central government command structure. AI infrastructure is commercial, multinational and fragmented. It serves civilian and creative purposes alongside defence and security applications.

The analogy explains the urgency. It does not accurately describe the ownership structure.

The AI race is being built with public resources

A private company may own the servers while still depending on systems, concessions and risks shared with the public.

• Power: generation, transmission, substations and grid connections
• Water: cooling supply, treatment systems and watershed capacity
• Land: zoning, roads and long-term development rights
• Networks: fibre routes and communications infrastructure
• Public support: grants, tax incentives, discounted services and expedited approvals
• Risk absorption: infrastructure that may remain if forecasts change or a project closes

These contributions are not automatically unjustified. A project can create construction work, skilled-trade opportunities, tax revenue, scientific capacity and domestic access to technology.

The stronger question is not whether a data centre creates any value. It is how much value, for whom, for how long and at what public cost.

Who pays: taxpayer, ratepayer or community?

Public cost does not appear through one channel.

• Taxpayers may fund grants, subsidies or tax concessions.
• Ratepayers may absorb part of the cost of utility upgrades through electricity bills.
• Municipalities may provide roads, water systems, fire protection or planning capacity.
• Communities may carry land, noise, water or environmental opportunity costs.
• Future customers may inherit long-term contracts and infrastructure commitments.

A project can receive no direct government cheque and still shift costs onto the public through utility infrastructure, tax concessions or long-term capacity commitments.

What the evidence supports—and what it does not

Confirmed

• Global data-centre electricity use is rising.
• AI is an important driver of projected future growth.
• Large-scale infrastructure expansion is underway.
• Governments increasingly treat computing capacity as strategic infrastructure.
• Local effects can be far larger than global averages suggest.

Supported analysis

• Compute concentration can increase corporate and national influence.
• Public infrastructure decisions can create long-term dependencies.
• Sovereignty claims depend on ownership, law, access and operating control.
• Infrastructure power can survive even if AGI predictions fail.

Not established

• Every facility raises household electricity rates.
• Every cooling system threatens local water security.
• Every community opposes construction.
• Every announced investment will be completed.
• One organization already controls advanced intelligence.
• AGI is inevitable or imminent.

Why a creator should care

Creators use the same infrastructure. Changes in computing cost, platform concentration and access can affect subscriptions, model availability, commercial terms and which tools survive.

That does not mean creators need to become utility engineers. It means that promoting a technology responsibly includes understanding enough about the system behind it to ask who controls the files, platform, audience and infrastructure.

Questions citizens should ask

Project purpose

• What will the facility operate?
• Who will use its computing capacity?
• Is it AI-specific, cloud, colocation or mixed use?

Electricity and water

• How much power has been requested, and when?
• What is peak demand compared with expected average use?
• Which cooling system and water source will be used?
• What is the difference between permitted, peak and expected water use?

Public cost

• Who pays for generation, transmission and substations?
• What tax incentives, grants or discounted rates were offered?
• Are public guarantees or long-term utility commitments involved?

Ownership and benefits

• Who owns the land, building, chips and computing capacity?
• How many temporary and permanent jobs are promised?
• What public access or local benefit is enforceable?

Long-term accountability

• What information must be reported after opening?
• What happens if the facility expands or changes owners?
• Who pays for stranded infrastructure if the project closes?

Who Controls the Intelligence?

This investigation continues across four connected reports:

1. The Race to Control AI Begins With the Data Centre
   Why the infrastructure race exists.
2. Who Pays to Power the AI Race?
   Electricity demand, grid construction and public cost.
3. The Water Behind Artificial Intelligence
   Cooling systems, measurement and local water concerns.
4. Public Money, Private Computing Power
   Subsidies, ownership, public return and long-term control.

The intelligence may be digital. The power behind it is physical.

AI infrastructure can create real public value. Governments and companies have legitimate reasons to expand computing capacity, strengthen domestic research and reduce dependence on foreign providers.

But citizens should not be asked to accept long-term commitments based only on broad promises about innovation, jobs or national leadership.

The power being built inside data centres will not remain inside the buildings. It will influence prices, research access, corporate competition, national dependence and the terms under which advanced systems are made available.

The intelligence may be digital, but the power behind it is physical—and the decisions made around that infrastructure will extend far beyond the buildings that contain it.

Further reading

This section contains Amazon.ca affiliate links. Jack Righteous may earn a commission from qualifying purchases at no additional cost to you.

• Chip War by Chris Miller — semiconductor supply chains and geopolitics
• Atlas of AI by Kate Crawford — AI’s material and institutional foundations
• The Coming Wave by Mustafa Suleyman and Michael Bhaskar — technological power and containment
• Empire of AI by Karen Hao — OpenAI and the institutions behind scaling

Primary sources

Energy and infrastructure

• International Energy Agency — Energy and AI

Compute concentration

• Stanford HAI — 2026 AI Index Report

United States

• OpenAI — Original Stargate announcement
• OpenAI — Stargate Michigan construction announcement

Canada

• Government of Canada — Canadian Sovereign AI Compute Strategy
• Government of Canada — National AI Strategy

Europe and the United Kingdom

• European Commission — AI Continent milestones
• UK Government — Compute Roadmap

Community reporting

• Brockovich Data Center Reporting

Continue exploring

Tech, Culture & Power Road
Investigations into AI infrastructure, media, public power and culture.

One Year After Sam Altman Declared the Singularity, Who Actually Benefited?
Why infrastructure power may be advancing faster than proven productivity.

The Righteous Beat
Selected creator, faith, technology and culture updates—not a daily news flood.