The data center conversation in Arizona has been framed almost entirely as a local question. Should this campus get permitted? Should that substation get approved? Is this water deal acceptable? Is that noise variance reasonable? All of these are legitimate questions and they deserve serious answers. They are also questions that take place inside a much larger frame that the local conversation almost never engages with.

The frame is this: the world is currently building, and will continue to build for at least the next decade, an enormous amount of new compute infrastructure. The infrastructure goes somewhere. The decision is not whether the buildout happens. The decision is which jurisdictions on the planet host it. That second-order question, where rather than whether, is the conversation Arizona is actually inside of, and the answers being chosen by other jurisdictions are worth understanding before deciding what kind of partner Arizona wants to be.

The Buildout Is Not Optional

A few facts that often get treated as up for debate, but aren't.

Global demand for compute keeps climbing, and AI workloads in particular have grown faster still, with training compute for frontier models scaling by orders of magnitude over the past several years. The infrastructure to host this demand has to exist. There is no path where the demand exists and the supply doesn't. The compute either gets built where it can, or it gets routed across the world to where it has been built.

This is not a question of whether AI is good or bad, useful or overhyped, dangerous or beneficial. The infrastructure is being built because the demand is real. Hospitals, banks, governments, retailers, manufacturers, transportation companies, military commands, and individual consumers are all using more compute every year. The trajectory was already steep before AI accelerated it. AI accelerated it more. Whatever happens with AI specifically, the underlying demand for compute will continue to grow for the foreseeable future.

So the question is not whether the buildout happens. The question is where, and under whose rules.

The Choices Other Jurisdictions Are Making

Several jurisdictions have already made their choices visible.

China has pursued a state-directed compute buildout under explicit national strategy. The "East Data, West Compute" initiative, launched in February 2022, routes data processing from the eastern cities to eight national computing hubs, several of them in western provinces with cheaper power (Inner Mongolia, Gansu, Ningxia, Guizhou). The state coordinates land, power, and water allocation. Environmental review is short. Local opposition is essentially absent. The compute that gets built is, by structure, available to the Chinese state for purposes the state determines, including security applications that Western governments would not permit.

Russia has announced and partially built sovereign compute infrastructure with explicit alignment to state surveillance and military applications. The scale is smaller than China's, but the alignment is similar.

Several authoritarian-leaning states in the Gulf (Saudi Arabia, the UAE) have announced multi-billion-dollar compute buildouts, often in partnership with Western companies, often with sovereignty arrangements that put the underlying infrastructure under host-state control. The labor arrangements that make these projects possible would not be permissible in the United States. The water and power arrangements would not be permissible in the United States. The buildouts are happening anyway.

European jurisdictions, particularly Ireland, the Netherlands, and the Nordic countries, have been important compute hubs but are running into infrastructure constraints. Ireland's grid operator effectively paused new data center connections in the greater Dublin area into the late 2020s, with data centers already drawing more than a fifth of the country's electricity. The Netherlands imposed a national moratorium on hyperscale builds in 2022 and moved siting under central control. Sweden and Finland have continued, but the available expansion is finite.

The major remaining democratic, rule-of-law jurisdictions where significant new buildout can plausibly happen at scale are the United States, Canada, Australia, and a handful of others. The United States is the largest of those by far, and the United States is currently absorbing the bulk of the global democratic-jurisdiction buildout.

What Sovereignty Actually Means in Practice

The phrase "compute sovereignty" gets used in policy circles. The practical version is concrete. It means the data center hosting the model that processes a CT scan, an immigration application, a tax filing, a court record, or a national security analysis sits in a jurisdiction whose courts and laws apply to what happens inside the building. It means the infrastructure cannot be commandeered by a foreign government. It means the workforce running the facility cannot be coerced, surveilled, or compromised by a state actor that is hostile to the user.

These are not abstract concerns. They are the explicit concerns of every major Western government's national security apparatus, and they are the reason the CHIPS and Science Act passed with bipartisan supermajorities and the reason the executive branch has been pushing domestic AI infrastructure across multiple administrations.

When a 100-megawatt data center gets built in Goodyear or Casa Grande, the campus is operating under United States law, Arizona water law, federal labor law, and the courts of a constitutional democracy. The workers have rights that they can enforce. The water rights are publicly recorded. The environmental review is documented. The community has access to public records and public hearings. The owner of the building cannot have it seized by a sovereign without due process. The data inside the building cannot be accessed by a foreign government without legal cooperation that the US government has to consent to.

If the same compute capacity gets built in a jurisdiction that doesn't offer those protections, the technology stack still functions, but the legal stack underneath it is different. The applications that run on top of it are different. The accountability is different. The recourse for individuals and institutions is different.

The decision to host significant compute in democratic jurisdictions is not a moral statement about other countries. It is a structural statement about what kind of infrastructure the democratic world wants to depend on for the next 30 years.

The Arizona Variant of the Question

Inside the United States, the question repeats at the state level. The compute will get built somewhere in the country. Which states host the buildout, and under what terms?

States that have made themselves friction-light have absorbed the bulk of the recent buildout. Virginia and Texas absorbed most of the buildout from 2010 to 2022. Both have run into capacity limits, in different ways. Virginia is grid-constrained and increasingly politically constrained. Texas has the grid (ERCOT) and the politics, but ERCOT's reliability has been compromised by extreme weather events and by the structure of the market itself.

Arizona, Georgia, Ohio, Oregon, the Carolinas, Iowa, and Indiana are absorbing the next wave. Each of these has different attributes. Arizona has the grid, the fiber, the climate-adjacent advantages, the workforce, the chip ecosystem, and the water management framework. Each of those is a real advantage. The combination of them is rare.

The states that don't absorb the buildout are not preventing the buildout from happening. They are choosing not to participate. The economic and infrastructure benefits flow to the participating states. The risks (whatever risks exist) are shouldered by the participating states. The trade is real, in both directions.

Arizona has already made the trade in part. The campuses are getting built. The fiscal benefits are flowing. The workforce is growing. The local conversations about water and power and noise are real local concerns that deserve real local answers.

The frame that's been missing is what the alternative would look like. The alternative is not "Arizona stays the same and the buildout doesn't happen." The alternative is "the buildout happens in Texas, or Georgia, or Ohio, and Arizona stays the same with a smaller industrial economy and a slower fiscal trajectory." The buildout itself is not negotiable. The participation is.

What's Worth Defending

The case for hosting compute in Arizona, and in democratic jurisdictions generally, comes down to a few things that the local conversation rarely articulates.

The compute is going to exist. The applications it powers are going to exist. The economic value it creates is going to exist. The geopolitical leverage it produces is going to exist.

The choice for any given community is whether to be inside the buildout or outside of it. Inside means accepting the construction phase, the substations, the truck traffic, the noise variances, the light pollution, the dust during construction, the changes to land use, the things that any large industrial project produces. Inside also means the property tax base, the construction employment, the workforce pipeline, the operational employment, the supply chain ecosystem, the bond capacity expansion, and the regional industrial diversification that makes the local economy more resilient.

Outside means none of those costs and none of those benefits. The compute moves elsewhere. The applications still get built. The economy still depends on them. The community is now a consumer of compute hosted somewhere else, with all the accountability and sovereignty implications that come with depending on infrastructure under someone else's control.

For a state to choose "outside" is a coherent choice. It is not a costless one. The conversation should at least be honest about what's being chosen on each side, because the framing that the buildout is optional has been the most consistently misleading element of the public discussion. The buildout is not optional. The location is. Once communities understand which question they are actually answering, the answers tend to look different from what the surface debate suggests.

Sources