Public Money, Private Computing Power

The Building May Be Private. The Deal May Be Public.

A proposed data centre often arrives through a large headline: billions in private investment, thousands of jobs, regional growth and a new place in the global AI economy.

Those promises may be real. They are not the complete transaction.

The less visible side of the agreement may include tax exemptions, discounted electricity, public transmission, roads, land preparation, worker training, faster planning, government-backed financing or direct grants.

Some of that support may create lasting public value. A road may serve several developments. A new substation may strengthen a region. A public supercomputer may remain available to researchers, startups and government institutions.

The largest number in the press release usually describes what the company says it may invest. It does not automatically reveal what the public is contributing or what citizens will retain.

Public Support Is Larger Than a Government Cheque

The word subsidy often makes people imagine one direct payment. A data-centre deal can involve several forms of support spread across different agencies and levels of government.

Not every public contribution is automatically improper. A new road can serve several businesses. A training program can create transferable skills. Public computing can widen access to a resource that would otherwise remain concentrated among the largest companies.

The correct test is not whether government participated. It is who primarily benefits, who carries the risk and who owns the asset after the agreement is complete.

The Argument for Public Investment

Advanced computing is becoming a strategic resource. Governments may support it to reduce dependence on foreign cloud providers, protect sensitive data, support science, strengthen national security and give domestic businesses access to the infrastructure needed to compete.

There is also a competition problem. Frontier AI requires chips, facilities and power at a scale that only a limited number of companies can afford. A purely private market may leave universities, startups, independent researchers and public institutions dependent on those companies.

Public investment can reserve capacity for researchers, smaller firms, safety testing, medical innovation or public-interest projects. That can be a stronger public case than reducing taxes for one privately controlled campus.

Public investment in compute is not inherently a gift to technology companies. Its value depends on who owns the infrastructure, who receives access and what happens when the equipment becomes obsolete.

“Thousands of Jobs” Can Mean Several Different Things

Employment is one of the most persuasive arguments used to support public assistance. It is also one of the easiest numbers to present without enough context.

Virginia’s Joint Legislative Audit and Review Commission found that a typical 250,000-square-foot data-centre building may have approximately 50 full-time workers, about half of whom are contractors. Construction of one building generally takes 12 to 18 months and may involve about 1,500 workers at its peak.

The same study estimated that the wider Virginia data-centre industry contributes 74,000 jobs, $5.5 billion in labour income and $9.1 billion in annual state GDP. Most of those economic benefits come from construction rather than ongoing operations.

Both supporters and critics can misuse these facts. A supporter can combine construction, permanent, contract and indirect work into one large number. A critic can count only people working inside the server building and ignore years of construction and supply-chain activity.

Review Virginia’s legislative data-centre study.

Public Return Is Larger Than Employment

Jobs are important, but they are not the only public benefit a community can receive.

Data centres may generate property tax, equipment tax, business rates, construction fees and local utility revenue. Those funds can support schools, housing, emergency services, public transportation, libraries or other community priorities.

An agreement can also reserve computing capacity for universities, startups or public agencies. It can fund apprenticeships, scholarships, local procurement or research partnerships.

Public infrastructure may remain after the original company leaves. Fibre, reclaimed-water systems, energy projects, land improvements or public computing can provide longer-term value.

A community-benefit promise is strongest when it is written, measurable, funded, independently audited and enforceable after a change in ownership.

Virginia: Large Economic Benefits and Large Tax Exemptions

Virginia is one of the world’s most established data-centre markets, making it useful for examining both economic contribution and public cost.

Virginia’s sales-tax exemption generally requires a qualifying data centre to make at least $150 million in new capital investment, create at least 50 qualifying jobs and pay those jobs at least one and a half times the prevailing local average wage. In designated enterprise zones or certain higher-unemployment locations, the threshold may fall to 25 jobs.

The exemption covers qualifying equipment and systems used to process, store and communicate data, including servers, networking equipment, chillers and backup generators.

Virginia’s legislative study estimated that the exemption provided approximately $928 million in tax savings during fiscal 2023 and was used by about 90% of the industry.

The same industry produces significant local revenue and statewide economic activity. That means the correct conclusion is not that the exemption created no value.

Citizens can still ask whether the minimum employment requirement remains appropriate as facilities become more automated, whether replacement equipment should remain exempt indefinitely and whether the investment would have occurred with a smaller incentive.

A 2026 Virginia report now requires aggregate information on the value of the tax benefit, investment, jobs and state and local revenue. Company-identifying details can remain protected, limiting the public’s ability to compare one recipient with another.

Review Virginia’s qualification rules and its 2026 aggregate reporting framework.

The United Kingdom Is Building More Than the Data Centre

The United Kingdom’s AI Growth Zone policy shows how public support can be assembled across electricity, planning, land, finance, skills and local taxation.

The government says the program could reduce the time needed to secure power by as much as five years. It also proposes targeted electricity support in selected regions where data-centre demand could reduce grid constraint costs.

The policy estimates that a qualifying 500-megawatt data centre could save up to £80 million per year in electricity costs. It projects up to £100 billion in additional investment and more than 10,000 jobs. These are government projections, not completed outcomes.

Local benefits include up to £5 million per Growth Zone for AI adoption and skills programs. Local authorities in England are expected to retain all additional business-rate growth for 25 years, with estimated revenue of approximately £5 million to £10 million per completed site each year.

By February 2026, the UK government said five designated zones were expected to create more than 15,000 jobs. When Parliament asked for the jobs to be separated into construction, permanent onsite, remote and indirect categories, the response did not provide a project-level breakdown and said no specific assumptions had been made about the geographic nature of indirect work.

That does not prove the jobs will not exist. It demonstrates why the category behind the headline matters.

Read the UK AI Growth Zone policy and the parliamentary jobs response.

Canada Is Funding Both Access and Ownership

Canada’s Sovereign AI Compute Strategy illustrates why public-compute programs should not be treated as identical to a tax break for a private data centre.

The strategy includes an AI Compute Access Fund of up to $300 million for Canadian small and medium-sized businesses. The program can cover two-thirds of eligible Canadian cloud-compute costs and half of eligible non-Canadian cloud-compute costs.

Funding may be non-repayable, repayable or conditionally repayable depending on the project’s expected public benefits and milestones. In May 2026, Canada announced support for 44 companies across sectors including health, energy, manufacturing, agriculture and transportation.

A separate AI Sovereign Compute Infrastructure Program provides approximately $890 million toward the design, construction and ongoing operation of a large-scale Canadian public computing system over seven fiscal years.

Canada is also discussing private large-scale data-centre projects. A May 2026 announcement involving TELUS said work was advancing, while explicitly stating that no funding had yet been committed or distributed.

These distinctions matter. Paying part of a small company’s cloud bill, constructing a Canadian-owned public supercomputer and negotiating support for a private campus create different ownership, access and risk arrangements.

Citizens should ask who selects users, what access will cost, how much capacity universities receive, whether public-interest research receives priority and what repayment the public receives when a funded commercial project succeeds.

Review Canada’s AI Compute Access Fund · public supercomputing program · TELUS infrastructure announcement

Europe Is Treating Compute Like Research Infrastructure

The European Union provides another public-investment model.

The European Commission says 19 AI Factories and 13 associated antennas are operational. The facilities connect AI-optimized supercomputers with universities, startups, small businesses, industry and public authorities.

Combined European and national investment in supercomputing infrastructure and AI Factories is expected to reach approximately €10 billion during the 2021–2027 period.

A separate InvestAI facility is intended to mobilize €20 billion for up to five AI Gigafactories capable of supporting next-generation models.

The public case is strongest when computing capacity is shared across researchers and smaller firms rather than controlled only by one subsidized company.

That model still raises questions. Who owns the processors? How is access priced? Which countries receive facilities? Can commercial customers receive priority? Who owns models trained with publicly supported computing? What happens when the hardware becomes obsolete?

Review Europe’s AI Factories and Gigafactory plans.

What Else Could the Public Have Funded?

Every public agreement has an opportunity cost. Money, land, electricity capacity and administrative attention committed to one project cannot be used in exactly the same way for another.

Possible alternatives include housing, hospitals, schools, public transportation, grid modernization, water infrastructure, manufacturing, scientific laboratories, public computing or smaller-business support.

A tax exemption is foregone revenue, not always a cheque that was already available for another program. It is still reasonable to ask whether the investment would occur without the exemption, whether a smaller incentive would produce the same result and what public asset remains when the agreement ends.

Opportunity cost does not prove that a project is a bad investment. It means the project should be compared with realistic alternatives rather than evaluated only against the promise of receiving nothing.

A Public Deal Should Create Public Protection

The recipient should not be the only source measuring whether the agreement succeeded. Independent audit and public reporting turn promises into evidence.

Could Stronger Requirements Drive Investment Elsewhere?

Developers argue that data centres can be built in competing jurisdictions. Slow approvals, public reporting and rigid employment conditions can raise costs or push projects into countries offering faster decisions.

Commercial confidentiality can be legitimate. Technology changes quickly. Permanent staffing may decline as automation improves. Construction and supply-chain effects are not captured by counting only workers inside the completed building.

These arguments deserve consideration.

Governments can compete through reliable electricity, skilled workers, strong universities, predictable taxation, clear rules and fast decisions. They do not need to compete only by offering the largest undisclosed subsidy.

Speed and accountability are not opposites. A government can provide a clear approval path while still requiring transparent costs and enforceable outcomes.

What We Know—and What Must Be Proven Deal by Deal

Creators Also Benefit From Public Investment

Creators use systems they did not build alone: public education, libraries, copyright law, electricity, internet infrastructure, university research and technologies developed through decades of public and private investment.

AI music also depends on semiconductor research, computing infrastructure and institutions supported by citizens.

I am not opposed to public investment in technology. Much of what creators use today exists because governments, universities, businesses and citizens shared the early cost.

That does not require accepting undisclosed agreements, weak job claims, permanent private control without public benefit or no consequences when promises fail.

Developing your public voice means learning to distinguish between investment that builds public capacity and a deal that merely transfers private risk.

Questions to Ask Before Public Support Is Approved

1. What is the complete value of the public support?
2. Does that include tax exemptions and utility support?
3. Which government agencies are participating?
4. What investment is announced, and what is contractually required?
5. How many jobs are construction jobs?
6. How many are permanent onsite jobs?
7. How many are contractors?
8. Which wages and benefits are required?
9. How many jobs must be filled locally?
10. What is the public cost per permanent job?
11. Which taxes will the facility pay?
12. Which taxes will it not pay?
13. How long do the exemptions last?
14. Are replacement purchases also exempt?
15. Who pays for roads, water, power and transmission?
16. Does the public own any resulting infrastructure?
17. Is computing access reserved for universities or local businesses?
18. What community benefits are guaranteed?
19. What information will be published annually?
20. Who independently audits the results?
21. What repayment is required if targets are missed?
22. Do obligations transfer to a new owner?
23. What happens if the facility closes early?
24. Is financial security provided for decommissioning?
25. Can citizens review the agreement before final approval?

Primary Sources Used

• Virginia Joint Legislative Audit and Review Commission: Data Centers in Virginia
• Virginia Department of Taxation: Data-Centre Sales-Tax Exemption
• Virginia 2026 Data-Centre Tax-Exemption Report
• United Kingdom: Delivering AI Growth Zones
• UK Parliament: AI Growth Zone Jobs Breakdown
• Canada: AI Compute Access Fund
• Canada: AI Sovereign Compute Infrastructure Program
• European Commission: AI Factories and Gigafactories

Public Support Should Produce Public Power

A data centre may be owned by a corporation, financed by investors and operated for private customers.

But when government supplies tax relief, electricity support, land, financing, planning authority or public infrastructure, the project is no longer a purely private transaction.

Citizens become participants in the deal.

That does not mean they should automatically reject it. It means they should be able to see the price, measure the return and enforce the promises.

The public contribution should create something the public can keep: revenue, employment, infrastructure, knowledge, access or strategic independence.

Otherwise, citizens may help build the computing power while receiving little control over what happens inside the data centre.