Tag: Government Accountability

  • The November 13 Ground Stop: What a Department of Technology Would Have Done Differently

    By the Founders, Department of Technology | San Diego, California | November 14, 2026

    On Friday, November 13, 2026, the FAA grounded flights at every U.S. airport for more than four hours. A future Department of Technology would not have run air traffic control, but we believe its structure would have made this failure less likely and the response faster and clearer. This companion piece explains how, and where our proposal has limits.

    The cause has not been confirmed, and the story is still developing. We rely on published reports and make clear where officials disagree.

    What we know so far

    The FAA said it issued a nationwide ground stop for all U.S. airports after communication issues affected multiple Air Route Traffic Control Centers. According to reporting from NBC News, instability was detected at 3:36 p.m. Eastern, the stop was issued at 3:54 p.m. and lifted at 8:15 p.m., and more than 7,000 flights were canceled. ABC News reported it was only the second nationwide ground stop since 9/11.

    Transportation Secretary Sean Duffy blamed old equipment. NPR quoted him saying, “We have old switches, old routers,” and that the system “got overloaded.” He also said the FAA never lost all radar or communications, and that the stop was a precaution.

    What we do not know

    Officials have not given one consistent explanation. CNN reported that Homeland Security Secretary Markwayne Mullin pointed instead to FAA updates added to the old system. Al Jazeera reported that investigators were examining whether a software update played a role. NPR noted that Duffy did not have a ready answer for why the systems became overwhelmed on an otherwise normal day, and that FAA leaders did not brief reporters. Some air traffic controllers also told reporters they do not accept the explanation, according to CNN.

    We do not claim to know which account is right, and the rest of this post does not depend on it.

    This was not a surprise

    Aging air traffic control technology has been a known problem for years. In January 2023, a failed FAA computer system led to a national ground stop. In 2025, repeated telecommunications failures hit Newark, and Duffy said at the time that “we could see other situations like this around the country because the system is old.” Those problems have spanned administrations of both parties. A known weakness that stays unfixed across years is a structural problem, not only a bad day.

    How a Department of Technology would help

    Our federal framework is a conceptual design, and a DoT would work alongside agencies like the FAA rather than replace them. Here is where we think it would have made a difference.

    1. Making the risk visible before a crisis

    The framework calls for a Technology Oversight Committee that audits operations and reports to Congress, plus regular public reporting. An audit of legacy network equipment across federal agencies would put a known weakness on the record, with a name attached and a deadline, long before it fails.

    2. Putting network expertise at the top

    The official explanation was “old switches, old routers.” Those are the exact tools network engineers work with every day. As we argue in Back to Basics, a leader with hands-on enterprise network experience can judge whether an upgrade plan is sound and whether a vendor’s promise is realistic, which is much harder for someone who has never run a production network.

    3. Testing changes before they go live

    If the outage did follow an update, as Mullin suggested, the lesson is about change management. Our framework already requires impact assessments for major technology initiatives. We would go further, with staged rollouts and tested rollback plans for critical systems. That last part is our suggestion and goes beyond the written framework.

    4. Planning for failure

    Grounding flights as a precaution was reasonable, and we are not second-guessing it. The goal is to shrink the time between failure and recovery through redundancy, monitoring, and rehearsed recovery plans. This is the thinking behind our work on continuity of government.

    5. Coordinating across agencies

    The framework says cross-agency technology initiatives should be coordinated through the DoT, and it creates a National Technology Security Council for sharing threat information. In the first hours of an outage, the public wants to know whether it is an attack. Shared incident protocols help officials answer that question quickly and consistently.

    6. Telling the public what happened

    Today, the public is hearing competing explanations. We would require a public post-incident report that explains what failed, why, and what will change. That fits our commitment to transparency and public reporting.

    The limits of our proposal

    A Department of Technology would not run air traffic control, and it could not replace aging equipment overnight. Modernization costs real money and takes years. We also cannot promise that it would have prevented this outage, since the cause is not yet known.

    Our claim is narrower: with accountable leaders, hands-on expertise, independent audits, and public reporting, a failure like this becomes less likely, and the response becomes faster and more transparent.

    Questions for officials and candidates

    1. Which critical systems in your agency still depend on legacy network equipment, and what is the replacement schedule?
    2. Who is personally accountable for the reliability of those systems?
    3. How are major upgrades tested, and how quickly can you roll one back?
    4. Will you commit to a public report after every major outage?

    The bottom line

    The ground stop was a warning, not just an inconvenience. Our airspace, like our schools, hospitals, and power grid, runs on networks that someone must be responsible for keeping up to date. A Department of Technology is our proposal for making sure someone is, and that voters can see how well they are doing.

    For the basics of our proposal, read What Is a Department of Technology? We will update this post as the investigation reveals more.

  • Self-Revising MI: Why Machine Intelligence That Rewrites Itself Needs a Plain-English Name

    By the Founders, Department of Technology | San Diego, California | November 14, 2026

    When a Machine Intelligence system can change its own code, its goals or its safeguards, what should we call it? MI (Machine Intelligence, aka AI or Artificial Intelligence) is the term we use throughout this site, and we explain why just below.

    The most common name for the self-changing kind is “recursive self-improvement,” and we think it is the wrong tool for the job. We propose a plainer one: Self-Revising MI.

    Recursive self-improvement AI or Self-Revising MI is and will be for the seeable future the most important, controversial, and civilizational altering aspect of MI in our lifetime.

    Whether you are for, against or undecided about SRMI (Self-Revising MI), a name matters more than it seems. Legislators, judges, regulators, reporters and voters all need to use the same words for the same thing. A good name has to be understood by a general reader on first hearing, and it has to hold up when it is written into a statute, a contract or an audit standard. Self-Revising MI meets both tests better than the alternatives we considered.

    A quick note: why we say MI, not AI

    We use MI, short for Machine Intelligence, instead of AI, short for Artificial Intelligence, because we believe it is the more accurate word. The intelligence these systems show is real, and it comes from a machine. “Artificial” suggests something fake or second-rate, which muddies how people think and talk about it.

    We also think the clearer word is better for people: for science, business, governance and law. It is the same idea behind this whole post. A plain, accurate name makes for clearer thinking and clearer rules, and we hope you will start using MI too.

    The problem with the current names

    “Recursive self-improvement” describes a specific, hypothesized scenario: a system improves itself, and the improved version is better at improving itself again, so progress could accelerate. That is an important idea, but it makes a poor category for rules. “Recursive” has a precise technical meaning in computer science (a procedure defined in terms of itself), and a system can change itself without any of that. “Improvement” assumes the change worked. A rule that only applies to systems that improve would force regulators to prove a success before they could act.

    Researchers use a crowd of overlapping labels. One 2026 survey of self-evolving agents sets its own boundary with three criteria: updates driven by experience, persistent effects, and self-initiated learning. A separate safety paper on self-evolving agent systems defines its scope with a different set of conditions.

    Earlier work on self-modification in rational agents uses yet another label. Each is useful inside its own field, but none gives a legislator or a judge a single, stable term.

    Our proposal, with a model definition

    Self-Revising MI is a Machine Intelligence system that is able to make persistent changes to its own code, model weights, training data or training process, objectives, tools or safeguards, without a person approving each individual change.

    Each part of that sentence is doing work.

    • “Self” means the system initiates or carries out the change, not an engineer.
    • “Revising” means a change to the system itself, not just to its output. Editing your own draft is the everyday picture.
    • “Persistent” excludes temporary adaptation, such as adjusting to the instructions in a single conversation. The 2026 survey above draws a similar line.
    • “Each individual change” means a one-time, blanket permission does not take a system out of the category. Approval has to be per change.

    This is an illustrative definition for discussion, not legal advice. Any real statute would need review by lawyers and technical experts.

    Four reasons “revising” is the right word

    • It describes a mechanism, not an outcome. A revision either happened or it did not. It can be logged, compared against the earlier version and reviewed. Oversight should not depend on arguing whether a change was an improvement.
    • It is plain English. Everyone knows what it means to revise something. No glossary is needed, which helps in a hearing room, a courtroom or a news story.
    • It is wide enough for how systems actually change. According to Arize, what changes in today’s self-improving agents is almost always context, prompts, retrieved examples or tool policy, while weight updates are rare. Meanwhile, the self-modification research notes that a system’s goals can change too. A definition that only covered retraining would miss most real cases. Naming the categories of change closes that gap.
    • It fits an oversight scale. One recent paper proposes grading systems by their revision authority, meaning the highest level at which a system can adapt itself without a human. That is the same idea we use below.

    Why not the other names

    • Self-Advancing MI assumes the change is progress. A rule tied to “advancing” invites a fight over whether any given change advanced anything.
    • Autogenic MI is precise but obscure, and the word already has other uses in other fields. It would need explaining every time.
    • Ouroboros MI is memorable, but it is a metaphor, and it sounds ominous. Metaphors do not make good legal categories, though it may survive as a nickname.
    • Bootstrap MI borrows from computing, where “bootstrap” already means several different things, including starting up a system and a statistical method.
    • Compounding MI describes a trajectory, not a capability. A system can revise itself without its gains ever compounding.
    • Self-Directed MI sounds like it is about goals and autonomy, and could be confused with systems that choose their own tasks but never change themselves.

    Keep “recursive self-improvement” for the scenario

    We are not asking anyone to retire the older phrase. “Recursive self-improvement” (or an “intelligence explosion”) can stay as the name of the runaway scenario that researchers worry about. Self-Revising MI is the observable, regulable capability that could lead there. Rules should attach to the capability we can see today, not to a scenario we are still debating.

    A three-level scale

    A clear name works best when it is part of a ladder. Here is a draft scale for discussion.

    1. Fixed MI. Changes only when people retrain or update it. Standard testing and change logs apply.
    2. Human-Approved MI. May propose changes to itself, but a named person approves each one. Proposals and approvals are logged.
    3. Self-Revising MI. Can change itself without per-change approval. Stronger safeguards apply: independent audits, tamper-resistant change logs, tested rollback, staged deployment and public reporting.

    This also matches how the technology is described today. Arize says the common shipping pattern is that automation detects problems and proposes fixes while a person approves them, which is Level 2 on our scale. And EM360Tech reports that OpenAI said in September 2026 that fully autonomous recursive self-improvement is not happening today, and that Anthropic says it is not there yet either, though both report that their systems are doing more of the work of building future systems. If that is accurate, the time to agree on vocabulary and ladders is before Level 3 becomes common, not after.

    What this means for a Department of Technology

    A Department of Technology would not decide how any individual system is built. Its job would be the accountable layer around it: keeping a shared vocabulary, requiring change logs, auditing the highest-risk systems and publishing what it finds. The same logic applied to the November 13 ground stop in our earlier post: when a critical system changes, someone should be able to say what changed, who approved it, and how to undo it.

    What we are asking

    • Use “Self-Revising MI” when you mean Machine Intelligence that can change itself without per-change human approval.
    • Ask officials and candidates to adopt a defined term, with categories of change named in the definition.
    • Ask developers and agencies to log every change a system makes to itself, and to say which level each system operates at.

    The bottom line

    The best names say what happened, not what we hope or fear will follow. Self-Revising MI tells you that a system changed itself, and that is the fact oversight needs first. It is clear to a general reader, specific enough for a statute, and leaves room for the older terms to keep their place. We will keep refining the definition, and we welcome your comments.

  • Fiber First? The Questions to Ask Before $50 Billion for Air Traffic Control

    By the Founders, Department of Technology | San Diego, California | November 14, 2026

    On Friday, November 13, 2026, the FAA grounded flights at every U.S. airport for more than four hours. In our first post, The November 13 Ground Stop: What a Department of Technology Would Have Done Differently, we looked at how accountable technology leadership could make a failure like this less likely. Since then, a natural question has come up: should we lay the fiber and the wiring first?

    Our short answer is yes, the network has to be the top priority. But “fiber first, everything else later” is not how a system that can never be switched off actually gets rebuilt. What matters even more than the order of the work is whether someone is accountable for the result.

    What has happened since Friday

    The cause of the outage has still not been confirmed, and the story is still developing. Here is what has been reported so far.

    • The funding request. Transportation Secretary Sean Duffy asked Congress for $50 billion in emergency funding to overhaul air traffic control, saying, “This is an emergency,” according to Bloomberg. He had asked for about $30 billion earlier in the day. The Standard reported that the earlier $30.1 billion request included $10.1 billion for air traffic control modernization, $2 billion for control towers, and $18 billion for airport projects.
    • The legislation. Duffy is urging passage of the Modern Skies Act, which would set aside $10 billion to finish upgrades to the digital architecture behind air traffic control, including telecommunications networks, according to ANI.
    • The money already approved. Congress approved $12.5 billion over five years in July 2025 for air traffic control upgrades and controller hiring, per The Jerusalem Post. Duffy says more is needed to finish the job. The Standard also reported that the administration wants action in the final weeks of the year, before the new Congress takes office in January.
    • Congressional oversight. Sen. Maria Cantwell said the Commerce Committee would review the outage, including the technology behind primary and backup communications frequencies, according to Implicator.
    • Competing explanations. Duffy blamed old, overloaded switches and routers. Our first post noted that Homeland Security Secretary Markwayne Mullin pointed instead to FAA updates added to the old system. Implicator reported that multiple sources told CBS News an update to the private network connecting the air traffic control system caused the failure. Officials have not reconciled these accounts.

    Why the network is the base layer

    It helps to think of air traffic control as three layers. At the bottom is the network: the copper lines, fiber, switches and routers that carry voice, radar and data between facilities. In the middle are the systems that use that data to track aircraft and keep them apart. At the top are the planning tools that decide when and where flights should go.

    That top layer now includes a new FAA scheduling and flow-management system called SMART, short for Strategic Management of Airspace, Routes and Trajectories. It is MI-powered, and MI (Machine Intelligence, aka AI or Artificial Intelligence) is the term we use throughout this site. According to the Department of Transportation fact sheet, SMART analyzes airline schedules, weather and airport capacity to predict congestion before it builds. It began limited operation around Washington, D.C. in September, and NPR reported that FAA and DOT leaders emphasize it is not a replacement for human air traffic controllers.

    Officials said Friday’s outage was not related to SMART, as the Taos News reported, and we have no reason to doubt that. The distinction matters for another reason, too. A planning tool cannot keep a network up. If the layer underneath fails, the smartest scheduling in the world does not help, because controllers cannot reach one another and planes stay on the ground. That is why we believe the backbone deserves first claim on attention and money.

    Why “fiber first” cannot be strictly sequential

    In our view, there are four practical reasons the work cannot simply be done in a neat order.

    • The system cannot be turned off. Air traffic control runs around the clock, so new connections have to be built alongside the old ones and cut over carefully. The FAA has said it expects to replace the remaining copper telecommunications wiring by September 2027, according to Travel Noire.
    • The projects are different workstreams. Scheduling software, towers, radar and telecommunications involve different vendors, contracts and timelines. Pausing one does not automatically speed up another.
    • How we change things matters. If the outage did follow a network update, as the CBS sources suggest, the lesson is that rollout practice matters as much as the hardware. New gear installed poorly can fail just as badly as old gear.
    • The scale is large. A Government Accountability Office review found that 51 of the FAA’s 138 air traffic control systems are unsustainable, and the FAA told GAO not to expect many modernization projects to be completed for at least a decade.

    Where reasonable people disagree

    Supporters of a large, up-front commitment argue that years of stop-and-start funding and shifting priorities are a core reason the system aged the way it did. Duffy has said the money is needed up front, as reported by KMPH. Skeptics point to the size of the request, the long history of delayed modernization projects, and the question of how the money will actually be tracked. Both concerns are fair, and they point to the same need: clear milestones and public reporting, whatever the final amount.

    Four accountability questions before a check is written

    1. Who owns the backbone? One accountable leader, with a name, should be responsible for the reliability of the communications network that air traffic control depends on.
    2. What are the milestones? A public schedule should say which facilities are being upgraded, by when, and how independent auditors will verify it. This is the kind of work our proposed Technology Oversight Committee is designed to do under our federal framework.
    3. How are changes tested and reversed? Critical systems should get staged rollouts and rehearsed rollback plans. That is our suggestion and goes beyond the written framework.
    4. What happens after an outage? Every major failure should produce a public post-incident report that explains what failed, why, and what will change.

    Where a Department of Technology fits, and where it does not

    A Department of Technology would not run air traffic control, write the funding bill, or replace equipment overnight. It would be the accountable layer around the work: auditing legacy systems across agencies, putting hands-on network expertise at the top, coordinating across agencies, and reporting to the public. We cannot say it would have prevented this outage, since the cause is not yet known. Our narrower claim, as we argued in Back to Basics, is that leaders who understand networks, and who answer to voters, are better placed to judge whether an upgrade plan is sound.

    Questions for officials and candidates

    1. Of the $50 billion requested, how much goes to the communications backbone, and how much to everything else?
    2. Which facilities still depend on legacy network equipment, and when will each one be replaced?
    3. Who is personally accountable, by name, for the reliability of that network?
    4. How are major updates tested, and how quickly can they be rolled back?
    5. Will you commit to a public report after every major outage?

    The bottom line

    Should we lay the fiber and the wiring first? Make it the first claim on money and attention, yes. But the larger lesson of November 13 is not about one layer of technology. It is that a known weakness stayed in place until it failed, and that no one could clearly say who was responsible for fixing it. Naming that person, publishing the schedule and reporting the results is how that changes.

    For the basics of our proposal, read What Is a Department of Technology? We will update this post as the investigation reveals more.

  • Back to Basics: Why Our Technology Leaders Must Know Networks

    By the Founders, Department of Technology | San Diego, California | November 14, 2026

    A future Department of Technology official should have real-world experience running enterprise-size networks: routers, switches, servers, and the websites that sit on top of them. Knowing a programming language or two is not enough, and neither is a university degree in computer science on its own. Hands-on network engineering experience is a must.

    This is our view, and we state it plainly. We are not claiming a survey or a study. We are describing what we see and why we think it matters.

    The Internet is made of physical things

    It is easy to think of the Internet as a cloud. It is not. It is cables, routers, switches, firewalls, and servers in buildings that need power and cooling, all configured by people who know how the pieces connect and what happens when one fails.

    A router decides where traffic goes. A switch connects devices inside a network. A server stores and delivers the websites, files, and services people use. When any of them is misconfigured or overloaded, services slow down or disappear, and the cause is rarely obvious to someone who has never worked with them.

    Machine Intelligence runs on networks

    The current Machine Intelligence (MI, often called AI) revolution and the Internet depend on each other. MI systems are trained and used across vast networks of servers in data centers, and the Internet is how those systems reach schools, hospitals, businesses, and homes. Better MI drives more demand for bandwidth and computing, and a stronger network makes better MI possible.

    That is why network engineers matter more than ever. The buildout we describe in our Data Centers series is not only about chips and algorithms. It is about power, fiber, cooling, and the networks that tie it all together.

    Programming skills alone are not enough

    Writing code is valuable, and many excellent technology workers are programmers. But being capable in one or a few programming languages is a different skill from designing, securing, and keeping an enterprise network running.

    A useful comparison is the road. Many people are skilled drivers, but we would not ask a skilled driver to design a highway interchange. Likewise, a talented programmer is not automatically qualified to oversee a statewide network or a county data center.

    In our experience, many self-described technology experts, including some in government jobs, have never configured a switch, built a routing table, or run a production web server. That gap matters most when they are asked to judge what a vendor is selling or what an agency actually needs.

    A computer science degree is not enough either

    This is not a criticism of computer science graduates. Computer science is a demanding and valuable field. But it is a different discipline from network engineering. In our view, a computer science degree typically emphasizes theory, algorithms, and software, while keeping an enterprise network running is a hands-on craft learned by doing it.

    In other words, for this job, a professional certification can be worth more than a generic computer science degree. Certifications such as Cisco’s and Juniper Networks’ networking credentials are built around the work itself: configuring routers and switches, securing traffic, and fixing what breaks. Vendor certifications in cloud computing from Amazon, Google, IBM, and others show the same practical, foundational skills. Many of them also expire, so holders must renew them as the technology changes.

    None of these replaces years of hands-on experience, and a public official should understand principles that apply across vendors rather than just one company’s products. But as evidence that someone has the skills a future Department of Technology official needs, a relevant certification on top of real-world experience says far more than a general degree does.

    Consider a comparison from the law. Someone with a philosophy degree has studied something worthwhile: logic, ethics, and argument. But if that person wanted to practice law, we would not call it practical or logical to let them represent clients on that degree alone. In nearly every state they would need law school, a bar exam, and real experience first.

    Technology leadership is no different. A degree in a related field does not, by itself, qualify someone to run the network that schools, hospitals, and emergency services depend on. The credential and the experience have to match the work.

    There is no technology equivalent of the bar exam, which makes a documented, hands-on track record more important, not less. A network engineer’s experience is a must for the people who lead and oversee public technology, not a nice extra.

    We already expect this of other infrastructure

    We rely on engineers to build and maintain our roads, bridges, water systems, and cities. No one would put someone who has never worked on a bridge in charge of bridge safety. Networks are infrastructure too, and our schools, hospitals, elections, emergency services, and economy increasingly depend on them.

    The Internet needs its own maintenance and its own engineers, and the people who lead public technology offices should come from that world or understand it deeply.

    What we would look for

    We would want a future Department of Technology official to bring real-world, hands-on experience in areas such as:

    • Enterprise networking: designing and running networks with routers and switches at a scale where outages affect many people.
    • Servers and hosting: deploying, securing, and maintaining web servers and the services that run on them.
    • Web building: creating and maintaining websites, not just consuming them.
    • Reliability and security: planning for backups, redundancy, monitoring, and recovery when something breaks.
    • Troubleshooting under pressure: diagnosing an outage in a live system, which no classroom fully teaches.

    Degrees and certifications can show that someone studied a subject. They cannot show that someone has kept a real network running, especially in a field that changes constantly. That is why we put a track record of hands-on work ahead of credentials alone.

    Why it matters in government

    Public technology decisions depend on this knowledge more than most people realize:

    • Data centers: officials weigh power, fiber, water, and cooling requests without always being able to tell a real project from a placeholder. See How a Data Center Deal Actually Gets Made.
    • Cybersecurity: defending public systems starts with understanding how they are connected.
    • Continuity of government: keeping services running through outages and disasters depends on resilient networks.
    • Procurement: leaders who understand the technology are harder to oversell and better at writing contracts.
    • Schools and public safety: systems like School Contact only work if the underlying networks are secure and reliable.

    What this does not mean

    A technology leader does not need to configure every router personally. Like any executive, they will rely on a professional staff, and we would fund one. But a leader who has never done the work cannot judge whether the staff, or a vendor, is giving good advice.

    Our first-year guide for counties and cities shows how much of the job comes down to asking the right questions. Leaders can only ask them if they understand the answers.

    Questions to ask candidates

    If someone is running for a technology role, or already holds one, here are a few simple questions voters can ask:

    1. Have you built or managed a network with routers and switches at enterprise scale?
    2. Have you run or secured a production web server?
    3. Can you explain, in plain language, how your agency’s systems connect and what happens if one fails?
    4. How will you make sure your team has the hands-on skills to defend and maintain our public networks?

    An honest answer, even one that admits a gap and explains how it will be filled, tells voters a great deal.

    The bottom line

    The Internet and Machine Intelligence are growing together, and both depend on engineers who know how networks really work. We maintain our roads and bridges because we know we need people who can build and fix them. We should hold the people who oversee our digital infrastructure to the same standard.

    We want to hear from network engineers, system administrators, and web builders. If you have run real networks, tell us what a public technology leader needs to know. We would rather be corrected than applauded.

  • Isn’t This Just Another Bureaucracy? Answering the Hardest Questions About Elected Technology Leaders

    By the Founders, Department of Technology | San Diego, California | November 13, 2026

    Our proposal for elected technology leaders at every level of government draws the same skeptical question again and again: aren’t you just adding a layer? It is a fair question. Here are the nine hardest objections we hear, with our best answers and the limits of those answers.

    We would rather be challenged than applauded. If an idea cannot survive its strongest critics, it should not become law. Where we have evidence, we cite it. Where we are making a judgment, we say so.

    1. Isn’t this just another layer of bureaucracy?

    It would be, if it added a layer without removing a gap. The gap is the problem. Today, the pieces of the technology buildout are split across utility commissions, economic development offices, zoning boards, treasurers and workforce agencies, and no one answers to voters for how they fit together. An NBC News review of more than 30 data center proposals in 14 states found that in a majority of cases local officials signed NDAs and worked with what appeared to be shell companies. That is not what a coordinated system looks like. Our test for any new office is whether it replaces confusion with accountability. If it does not, it should not exist. We also think a new office should come with a published cost estimate and a sunset review. We have not priced it for your jurisdiction, and you should be wary of anyone who claims to have done so without details.

    2. Why elected? Why not appoint someone?

    Because the decisions are value judgments. Whose land, whose power bills and whose jobs are tradeoffs, not technical questions, and voters should be able to hold someone to account for them. There is precedent. Voters elect utility commissioners in 10 states, and on November 3 voters in nine states elected 14 members of their state utility commissions. That is not an untested idea. But the record also contains a warning: those races draw far less public attention than other contests on the same ballot, even though the winners will help decide whether data center costs are passed to existing ratepayers. And S&P’s regulatory research unit has long attributed greater investor risk to states where commissioners are elected. So election alone does not guarantee accountability. It needs visibility, and we would pair it with the transparency rules we describe elsewhere.

    3. Don’t utility commissions and other regulators already do this?

    They do part of it, and in some places they do it well. Virginia’s State Corporation Commission, for example, ordered Dominion to develop a tariff to assign more transmission costs directly to data centers. But a utility commission sets rates. It does not decide where a facility is zoned, how much water it may use, what tax breaks it receives or how a county prepares its budget for a downturn. The proposal is not to replace regulators. It is for a technology leader to participate in their proceedings, and to coordinate across the decisions that no single regulator owns.

    4. Won’t an elected technology leader be captured by the industry?

    It is a real risk, and it is a risk for any regulator, elected or appointed. Campaign money, revolving doors and information asymmetry all push in one direction. We would counter it with rules, not hope: publish meetings and calendars, bar NDAs, require recusal where there is a financial interest, and open the records. Those safeguards are our suggestions, not a guarantee. A reasonable person can conclude that appointed, professional regulators are safer than elected ones, and that is a debate we welcome.

    5. Will they have the expertise?

    Not automatically. Elections reward name recognition more than engineering. Some states already set minimum qualifications for commissioners; a Platte Institute research note observes that some states require knowledge or experience in fields such as economics, finance, engineering, energy and law. We would consider something similar, along with a funded professional staff, so that the elected official is accountable for the decisions and the staff supplies the analysis.

    6. Won’t this slow down the buildout and hand China the lead?

    It should not, and if it does, the design has failed. Speed matters. Our concern is that speed without accountability produces backlash that stops projects anyway, as voters showed on November 3 when they favored limits in 40 of 41 local data center measures. Clear rules can also speed things up: AEP Ohio’s large-load pipeline reportedly fell from roughly 30 GW to 13 GW once it had a tariff, which suggests that rules can weed out speculative requests that waste everyone’s time. And Loudoun County credits favorable zoning and fast approvals for accelerating its data center growth. A good technology leader should make the process predictable and fast for projects that meet the standards.

    7. What will it cost?

    We do not have a number for your jurisdiction, and we would not trust one without a detailed budget. The fair comparison is with the cost of the current arrangement, though that is not proof of savings. In Ohio, for example, Policy Matters Ohio reports that a single tax break cost state and local governments about $2 billion in a year, well above the state’s forecast. A small county could begin with a staff director and a part-time advisory board and expand only if it proves its worth.

    8. Isn’t this a partisan project?

    It is not meant to be, and the evidence we rely on is bipartisan. In Virginia, 21 Democrats and seven Republicans in the state Senate voted to curtail a data center tax break. Republican Governor Mike DeWine paused Ohio’s break, and Democratic Governor Andy Beshear called for ending Kentucky’s. Skeptics of the buildout and enthusiasts for it both need an accountable referee. We will count the proposal a failure if one party sees it as a weapon.

    9. Why not let states and markets sort it out?

    They will sort out something. In 2026, 29 states enacted AI legislation, and many local officials are negotiating one deal at a time against a handful of very large buyers, often under NDAs. A patchwork can be a healthy laboratory, and a federal standard may eventually be the right answer. Either way, the people who live next to the project need someone whose job is to speak for them across every decision, not just the one in front of a given committee.

    What would change our mind

    We care about the function more than the title. If a jurisdiction shows that a staff-level technology director, a joint committee or an existing regulator can deliver one accountable owner for the whole picture, with public records and a name voters can hold responsible, we would support that model. If evidence shows elected technology leaders performing worse than appointed ones, we will say so. What we will not accept is the status quo, in which nobody is responsible. Our proposal is at department.technology/.

  • The Data Center Casino: Why America Needs Elected Technology Leaders Now

    By the Founders, Department of Technology | San Diego, California | November 13, 2026

    America needs more data centers. That part is not in dispute. What is in dispute is who decides where they go, who pays for them, and whether anyone is looking at the whole board.

    If the United States wants to remain the world leader in artificial intelligence, it has to build the computing capacity, and the power to run it. We start from that premise. But for several years, officials in both major parties, in red states and blue, have treated new data centers less like public infrastructure and more like a slot machine: pull the lever, hope for a jackpot of investment, tax revenue and ribbon-cuttings, and worry later about who covers the losses.

    The question that has been asked is “how much money can be made?” The question that should be asked is “what is the best outcome for the community that has to live next to this?”

    Three risks pull in three different directions. Build too fast, and we strain an aging grid and inflate a financial bubble. Build too slow, and other countries may make our lead obsolete. Ignore what AI could do to employment, and we erode the tax base a modern society runs on. No one in government is currently accountable for all three at once.

    The casino logic

    Thirty-eight states offer tax incentives to data centers, and at least 28 of them considered scaling those incentives back this year, according to the National Conference of State Legislatures. At least ten states forgo more than $100 million a year in these breaks, with Texas and Virginia each losing as much as $1 billion annually, according to Good Jobs First, as reported by Broadband Breakfast.

    In Ohio, Policy Matters Ohio reports the break cost state and local governments about $2 billion last year, far more than the state had forecast, which led Republican Governor Mike DeWine, who had previously vetoed a repeal, to order a temporary pause.

    The backpedaling is bipartisan, which is exactly the point. In Virginia, 21 Democrats and seven Republicans in the state Senate voted to curtail a 5.3% tax break. Kentucky’s Democratic governor, Andy Beshear, has called for ending his state’s incentives, and Ohio, Arizona, Nebraska and Illinois have paused theirs.

    When lawmakers of both parties are reversing course, the original approach was not a partisan mistake. It was a structural one: the deals were made one project at a time, often in secret (as we noted in our open letter, 25 of 31 Virginia local governments involved in data center development had signed NDAs), and no one owned the cumulative consequences.

    Build too fast: the grid and the bubble

    Our electrical grid is where the cost of speed shows up first. In PJM, the nation’s largest grid market, wholesale power prices jumped almost 76% in the first quarter of 2026 compared with a year earlier, and the market monitor said the price effects on customers were very large and not reversible, according to E&E News.

    Capacity auction prices rose roughly tenfold in two years, from about $29 to $329 per megawatt-day, and the December 2025 auction fell short of its reliability target for the first time, per one analysis of PJM’s results. Ordinary ratepayers are paying for capacity built to serve some of the wealthiest companies in the world.

    Then there is the money behind the buildout. Hyperscalers will spend about $750 billion this year, and they are borrowing heavily to do it, while economy-wide statistics still show little or no productivity growth from AI. Jefferies warns that AI investment was the main driver of U.S. economic growth last year after consumer spending, so a sudden collapse would leave the economy exposed. And The Economist notes that more than half the spending has gone to servers and specialized chips that become obsolete within a few years. A technology can be transformative and still be overbuilt. If that happens, the stranded assets and the unpaid debts will not stay on Wall Street.

    Build too slow: the electron gap

    The opposite risk is just as real. The United States leads in advanced chips but is short on power, while China has the power and is short on chips. BloombergNEF estimates China will add more than six times as much generating capacity as the U.S. over the next five years. Morgan Stanley has forecast a 44-gigawatt shortfall for U.S. data centers within three years, roughly the summertime capacity of New York State. According to calculations reported by The Washington Post, Chinese data centers pay about a third of what American ones do for electricity.

    We cannot answer that gap with moratoriums and lawsuits alone, and we cannot answer it with secrecy and subsidies either. Speed that costs public trust is not speed. As voters showed on November 3, when they favored limits in 40 of 41 local data center measures, communities will stop projects they feel were done to them rather than with them.

    Ignore the jobs: the tax base

    The third risk is the quietest. A RAND analysis found that 84 percent of federal revenue in 2024 came from individual and payroll taxes, both collected from labor income. Because income is concentrated in high-paid white-collar jobs, replacing those roles with AI would hit revenue harder than the headcount suggests, and higher corporate profits would not make up the difference unless corporate tax rates roughly doubled.

    The same logic applies closer to home. New York City’s Comptroller built five AI scenarios for the city’s finances, put a combined 50 percent probability on the three with worse outcomes for jobs, growth or tax revenue, and recommended building a reserve fund to 16 percent of annual tax revenue. Every county and state should be asking the same question before the layoffs arrive, not after.

    Why we need elected technology leaders

    Look at who currently holds a piece of this problem. Utility commissions watch rates. Economic development offices watch ribbon-cuttings. Treasurers watch revenue. Each sees one slice, and none is answerable to voters for the combination. That is how a state can win a data center, lose a billion dollars a year, raise its residents’ power bills and weaken its future tax base, all through decisions that each looked reasonable on its own.

    We are calling for elected technology leaders at the local, county, state and federal levels, with one job: to weigh speed, grid capacity, public cost and long-term fiscal health together, in public. In practice that means:

    • Public plans before approvals. Grid, water and fiscal impact published before a project is approved, not after.
    • No secret deals. Public bodies should not sign NDAs that keep residents from learning what their own governments agreed to.
    • Large users pay for what they trigger. Grid upgrades and capacity costs assigned to the facilities that cause them, not spread across households.
    • Incentives with accountability. Tax breaks tied to measurable jobs and revenue, with clawbacks if the promises are not met.
    • Fiscal readiness for AI displacement. Scenario planning and reserves at every level of government.

    They should be elected, not appointed, because the tradeoffs here are value judgments about whose power bills, whose land and whose jobs, and those belong to voters. And they are needed now because the contracts are being signed now.

    The case against our case

    A fair reading includes the strongest objections.

    • Incentives may be what brings projects at all. The Data Center Coalition’s Dan Diorio argues that revenue estimates overlook the chance a project would never have been built without exemptions, and notes that states with the most growth, including Virginia, Texas and Pennsylvania, offer them, while Colorado, which does not, has less (S&P Global).
    • The electron gap may be overstated. The Brookings Institution cautions that no one knows how many data centers will be needed or how fast chips will become more efficient, and that worries from builders are not the same as geopolitical shocks (Brookings).
    • Job forecasts are uncertain. One UK think tank’s scenarios range from eight million jobs lost to none at all (IPPR), and New York’s Comptroller itself says it is hard to tie current trends to AI with certainty.
    • Another elected office is another layer. Critics may reasonably prefer strengthening utility regulators and legislatures over creating new positions. Our answer is that each existing body owns only a slice, but that is a debate worth having in public.

    Start the conversation

    Neither the casino nor the moratorium is a plan. The country needs people whose job it is to build enough, build in the right places, make the builders pay their share, and prepare for what comes after. That job does not exist yet. We are asking every candidate and every sitting official, of every party, to start the discussion about creating it, at department.technology/.

  • Ten Questions Every Candidate and Elected Official Should Answer About Data Centers and AI

    By the Founders, Department of Technology | San Diego, California | November 13, 2026

    Voters deserve specific answers about who pays for data centers, who is accountable, and what happens if AI does not deliver. Here are ten questions you can put to any candidate or sitting official, with the evidence behind each, what a strong answer sounds like, and what evasion looks like.

    This is a nonpartisan tool. We score the quality of an answer, meaning whether it is specific, honest about tradeoffs and accountable, and not whether it matches our views. A candidate who answers concretely in favor of a policy we would not choose earns the same points as one who answers concretely against it. Send the same ten questions to every candidate, give everyone the same deadline, and publish the answers word for word.

    How to score

    For each question, give each answer 0, 1 or 2 points:

    • 2 points: a clear position, a specific mechanism, and an honest acknowledgment of a tradeoff or limit.
    • 1 point: a clear position without a mechanism, or a mechanism without a position.
    • 0 points: no answer, a deflection, or generalities that could mean anything.

    A total of 16 to 20 suggests a candidate who has thought hard about the whole picture. A total of 10 to 15 suggests a developing view. Below 10 suggests the issue has not yet been taken seriously. This is a conversation tool, not a scientific rating, and we encourage independent groups, newsrooms and civic organizations to score the answers themselves.

    Question 1: Who is accountable?

    The question: If a data center proposal comes before your body, who in government is responsible for weighing power, water, land, taxes and jobs together? Who should be?

    Why it matters: The pieces of this decision are split among utility commissions, economic development offices, zoning boards and treasurers, and no one answers to voters for the combination.

    A strong answer names a person or office and a mechanism, such as a standing committee, a technology director or an elected leader, and says how voters can hold them to account.

    Red flags: “The market will sort it out,” or an answer that lists existing agencies without saying who coordinates them.

    Question 2: Will you sign secret agreements?

    The question: Will you sign a nondisclosure agreement about a development project, and will you support limits on them?

    Why it matters: An NBC News review of more than 30 data center proposals in 14 states found that in a majority of cases local officials signed NDAs. Records requests in Louisiana turned up 54 NDAs signed by elected officials since early 2024. Microsoft says it is working to end active NDAs governments have signed.

    A strong answer gives a clear yes or no, says when confidentiality ends (for example, before any public vote), and explains how legitimate trade secrets would be protected.

    Red flags: “NDAs are standard practice” with no limit, or a blanket promise with no mention of how negotiations would work.

    Question 3: Who pays for the grid?

    The question: Should the largest new electricity users pay the full cost of the grid upgrades built to serve them? Should that be mandatory or optional, and what minimum commitments should they make?

    Why it matters: The Ratepayer Protection Act passed the House 417 to 3 and then failed a Senate procedural vote 57 to 43, largely over whether the protection should be optional or mandatory. At the state level, 23 states had approved at least one large-load tariff as of May. Ohio’s AEP tariff requires 85 percent minimum payments and at least twelve-year contracts, and Dominion’s Virginia tariff requires 14-year contracts and $1.5 million per megawatt of collateral.

    A strong answer takes a side on mandatory versus optional and names at least one term, such as minimum bills, contract length, collateral or exit fees.

    Red flags: “Make Big Tech pay their fair share” with no mechanism, or “markets will handle it.”

    Question 4: Do you support tax breaks, and on what conditions?

    The question: Do you support tax incentives for data centers? If so, what conditions, such as job targets, sunset dates and clawbacks, should come with them? If not, why not?

    Why it matters: Thirty-eight states offer data center incentives, and at least 28 considered scaling them back this year. In Ohio, the break cost state and local governments about $2 billion in a year. The jobs are mostly temporary: an Illinois study projected 121,000 jobs from $57 billion of investment, but only 2,800 permanent direct positions, while a Reno facility’s records estimated 73 permanent jobs.

    A strong answer takes a position either way, separates temporary from permanent jobs, and says how results would be measured.

    Red flags: Quoting job totals without saying how many are permanent, or promising “jobs and growth” without numbers.

    Question 5: Who gets a say before approval?

    The question: Should data centers be allowed by right, or should each one require a public hearing and discretionary approval? Where should the final decision sit?

    Why it matters: Loudoun County eliminated by-right zoning for data centers, and Harrisonburg, Virginia, moved them to special-use permits. Loudoun also credits favorable zoning and fast approvals for accelerating its growth. There is a real tradeoff between speed and consent.

    A strong answer describes a process, such as hearings, timelines and standards, and acknowledges the tradeoff.

    Red flags: “Let the community decide” with no process, or “get out of the way” with no protection for residents.

    Question 6: What about water, land and cleanup?

    The question: What studies, standards and financial guarantees should a data center have to meet for water, noise, farmland and eventual closure?

    Why it matters: Linn County, Iowa, requires a water study and a water-use agreement before construction and also covers decommissioning bonds. Few places have anything comparable.

    A strong answer names specific requirements and who would enforce them.

    Red flags: “Environmentally responsible” with no specifics.

    Question 7: How do we build enough, fast enough?

    The question: How would you make sure America builds enough power and computing capacity to compete, without bypassing the public?

    Why it matters: BloombergNEF expects China to add more than six times as much generating capacity as the United States over five years, and Morgan Stanley has forecast a 44-gigawatt shortfall for U.S. data centers. At the same time, AEP Ohio’s pipeline reportedly fell from roughly 30 GW to 13 GW once it had a tariff, so not every request is real.

    A strong answer addresses generation, transmission and permitting, and distinguishes real projects from speculative ones.

    Red flags: An answer that is only “beat China” or only “stop the data centers.”

    Question 8: Who bears the loss if the boom slows?

    The question: If AI investment slows or reverses, who bears the losses, and what would you do now so residents are not left holding the bag?

    Why it matters: Hyperscalers will spend about $750 billion this year, borrowing heavily. Jefferies warns that AI investment was the main driver of U.S. economic growth last year after consumer spending, and The Economist notes that more than half the spending goes to equipment that becomes obsolete within a few years.

    A strong answer names protections such as collateral, minimum bills, exit fees or cleanup bonds, without predicting a crash.

    Red flags: Either “this can’t fail” or doom with no plan.

    Question 9: What is your plan for the tax base?

    The question: Has your government modeled what happens to revenue if AI displaces significant work, or if data center revenue falls? What reserves and triggers are in place?

    Why it matters: A RAND analysis found that 84 percent of federal revenue comes from individual and payroll taxes. Loudoun County reports that data centers generate 38 percent of its general fund revenue, and New York City’s Comptroller recommended a reserve of 16 percent of annual tax revenue in response to AI risk.

    A strong answer says what share of revenue is exposed and describes a scenario plan and reserve.

    Red flags: “We’ll cross that bridge when we come to it.”

    Question 10: Should Congress override state AI laws?

    The question: Should federal law preempt state AI rules? If so, what federal protections and enforcement would replace them, and who would administer them? (For state and local candidates: what role should your jurisdiction keep?)

    Why it matters: The Senate voted 99 to 1 against an early preemption proposal, while 29 states enacted AI legislation in 2026. Even the White House’s approach would leave data center infrastructure to the states.

    A strong answer takes a position, names the enforcement mechanism and says who is accountable for it.

    Red flags: “Washington can’t be trusted” or “states can’t be trusted” with no alternative.

    Which questions fit which office

    • City council or county board: Questions 1, 2, 5, 6 and 9.
    • State legislature or governor: Questions 2, 3, 4, 7, 8 and 10.
    • Congress: Questions 3, 7, 8, 9 and 10.
    • Bonus question for everyone: Will you publish your meetings with developers, utilities and lobbyists on this issue?

    The case against our case

    • Scorecards reward rehearsed answers. A well-coached candidate can sound specific without meaning it. That is why we ask for mechanisms and why we suggest publishing answers in full and comparing them with later votes.
    • Not every office has the power. A county commissioner cannot set federal policy, and a member of Congress cannot rezone land. The “which office” guide above is meant to keep the questions fair.
    • Ten questions oversimplify. Real issues have more dimensions, and some candidates will rightly say “it depends.” We would count a thoughtful “it depends”, with the conditions named, as a strong answer.
    • We might be seen as scoring for our side. Our proposal for elected technology leaders is not the only acceptable answer to Question 1, and a different structure can score full points. We would prefer that a neutral organization do the scoring.

    Print-ready list

    1. Who in government is responsible for weighing power, water, land, taxes and jobs together, and who should be?
    2. Will you sign a nondisclosure agreement about a development project, and will you support limits on them?
    3. Should the largest new electricity users pay the full cost of grid upgrades built to serve them, and should that be mandatory or optional?
    4. Do you support tax incentives for data centers, and on what conditions?
    5. Should data centers be allowed by right, or require a public hearing and discretionary approval?
    6. What studies, standards and guarantees should a data center meet for water, noise, land and cleanup?
    7. How would you make sure America builds enough power and capacity to compete, without bypassing the public?
    8. If AI investment slows, who bears the losses, and what would you do now to protect residents?
    9. Has your government modeled what happens to revenue if AI displaces work or data center revenue falls?
    10. Should Congress override state AI laws, and if so, what replaces them and who administers it?

    Use it

    Send these to every candidate and every sitting official you can reach, at every level, and ask for answers in writing by a stated date. Publish what comes back. If you are a candidate, answer before you are asked. Our proposal is at department.technology/.

  • How a Data Center Deal Actually Gets Made, and Where the Public Gets Left Out

    By the Founders, Department of Technology | San Diego, California | November 13, 2026

    Most data center deals are shaped long before residents hear a word. Here is a step-by-step walk through how a typical one comes together, who decides at each stage, where the public is shut out, and what an accountable technology leader would change.

    No two deals are identical, and we do not claim every project follows this path. Some officials negotiate openly, and some developers welcome scrutiny. But the pattern below appears again and again in public records and investigations, and understanding it is the first step to changing it.

    Step 1: A site is chosen, and it is not about you

    Data center developers pick locations for electricity access, water rights and fiber connectivity, not for population or local jobs, according to a Quartz analysis of the industry’s land-acquisition playbook. Your community may be on a list for reasons no one has told you.

    Who decides: the developer. Where the public is left out: everywhere; no one has been told. What would change: a technology leader would know the utility’s pipeline and whether your community is on it (see our first-year guide).

    Step 2: Land is quietly secured

    The next move is usually to lock up land. The most common tools are, according to that same analysis, shell LLCs, nondisclosure agreements, land options that tie up parcels before public review, and intermediary buyers acting for unnamed principals. The practice is old. In 2019, the Washington Post reported that Google had used shell companies in negotiations with at least five cities, and in Iowa created one LLC for the land sale and another for the development deal.

    There is a legitimate reason for some of this. Once a famous buyer is identified, land prices can jump sharply; the article cites Disney in Florida, where prices rose more than 1,000 percent. Sellers deserve a fair price, and developers deserve a fair negotiation. The question is how long secrecy should last and who else is kept out.

    Step 3: Officials sign NDAs, and projects get code names

    Local officials are then asked to sign confidentiality agreements. An NBC News review of more than 30 data center proposals across 14 states found that in most cases local officials signed NDAs and dealt with what appeared to be shell companies. In Louisiana, public records requests turned up 54 NDAs signed by elected officials since early 2024, and the agreements covered projects under code names like “Project Lightning.” In Minnesota, a project near Hermantown, code-named “Project Loon,” was identified as a data center only after the Star Tribune obtained public records, by which point the city had already changed land use plans and carried out preliminary environmental review.

    Who decides: a handful of local and state economic development officials. Where the public is left out: the officials who sign cannot discuss it, even with their own colleagues or constituents. What would change: a policy that the jurisdiction does not sign NDAs on development matters, with trade secrets protected by narrower means.

    Step 4: The utility is asked for power

    In parallel, the developer asks the utility for a connection. AEP Ohio, for example, charges a one-time fee of $10,000 to $100,000 for each load study and requires it to be paid within 45 days or the customer forfeits its place. These queues can be full of placeholders. AEP Ohio was facing 30 gigawatts of potential new load before its data center tariff, and one analysis found that its pipeline fell from roughly 30 GW to 13 GW after the tariff took effect.

    Who decides: the utility, under rules set by the state commission. Where the public is left out: the pipeline is rarely published, so local officials may not know which requests are real. What would change: public reporting of large-load requests by jurisdiction, and a local seat in the commission proceeding.

    Step 5: Incentives are negotiated

    The developer then seeks tax relief. In the Midlothian, Texas, case the Post described, Google reportedly used a shell company to negotiate a $10 million tax incentive with local officials. At the state level, thirty-eight states offer incentives, and in Ohio the break cost state and local governments about $2 billion last year, according to Policy Matters Ohio, far more than forecast. Those figures come from the incentive programs themselves, not from one negotiation, but they show the scale of what is decided in small rooms.

    Who decides: state and local economic development officials. Where the public is left out: the cost-benefit analysis, if one exists, is often not public. What would change: published cost-benefit analyses before approval, and clawbacks if promises are missed.

    Step 6: Land use is approved

    Only now does the public typically see a rezoning request or a site plan. If data centers are allowed by right, there may be no hearing at all. By this stage, the land is secured, the utility has been approached and the incentives are drafted, so the room to say no is small. In Hermantown, by one account, officials turned away questions from nearby residents about whether the project was a data center.

    Who decides: the planning commission and the county board or council. Where the public is left out: the decision arrives late and often looks finished. What would change: special-use permits instead of by-right approval, and a requirement for a public hearing with full disclosure before anything is signed.

    Step 7: The power contract is signed

    The utility and the developer sign a service agreement. In states with large-load tariffs, these carry minimum bills, long terms and collateral. Dominion’s Virginia tariff, for example, calls for 14-year minimum contracts and $1.5 million per megawatt of collateral. Elsewhere, the terms are weaker or negotiated one at a time. Who bears the risk if the facility closes or never fills is a question of contract design, and most residents never see the contract.

    Step 8: Construction, jobs and revenue

    Then the buildings go up. The jobs promised are mostly short-term. A Reno facility’s business records estimated 73 permanent jobs against more than 4,000 temporary construction jobs. An Illinois study projected that $57 billion of new data center investment would create 121,000 jobs, but only 2,800 would be direct, permanent positions, or about $20 million of investment per permanent job. Brookings researchers found effects of roughly 100 to 200 jobs at a typical affected county, along with a modest increase in home prices of 2 to 5 percent, and noted the effects vary by facility type.

    On the other side of the ledger, the tax revenue can be large. Loudoun County, Virginia, reports that data centers produce 38 percent of its general fund revenue on about 4 percent of its commercial land.

    The pattern

    Laid end to end, the steps show a consistent shape: the biggest decisions happen early, in private and without a public voice, and the public hearing arrives after the key commitments have been made. That is not necessarily bad faith. Many officials believe they are protecting a competitive negotiation, and many developers are responsive once the project is public. But a system that depends on goodwill at every step will not protect residents when goodwill runs out.

    The case against our case

    • Officials say NDAs are necessary. Louisiana’s economic development secretary told NPR that NDAs are “fundamental for economic development”, allowing the state to compete. Her point deserves weight. On the other hand, a parish president in the same state said economic development does not require NDAs, chose not to use them, and still saw a $10 billion data center built.
    • Early secrecy can protect sellers. As noted above, a named buyer can inflate land prices, so a limited period of confidentiality may serve landowners. We would limit it to land transactions and end it before any public vote.
    • The jobs numbers are disputed. A Maryland Tech Council study, for example, estimated that a typical 800,000-square-foot data center would support nearly 500 permanent jobs, a far higher figure than other studies, according to one review of the evidence, which also cites a Virginia state audit that found far fewer.
    • Many deals are fine. Some projects are welcomed by communities and deliver what they promise. The goal is not to stop deals but to make sure the public can see them before they close.

    What to do with this

    If you serve on a council or board, ask where your community is in this sequence, and who has signed what. If you are a resident, request the NDAs and incentive agreements under your state’s public records law. And if you are a candidate, tell voters who in your government is responsible for each of these eight steps. Right now, the honest answer in most places is that nobody is responsible for all of them. Our proposal is at department.technology/.

  • What Every County Treasurer Should Model Before AI Layoffs

    By the Founders, Department of Technology | San Diego, California | November 13, 2026

    Local governments face two different AI-related revenue risks at once: the labor income that funds them could shrink, and the data center revenue that may be replacing it could be concentrated in one fragile industry. Here is a simple scenario worksheet any treasurer or finance director can build in an afternoon.

    This is not a forecast. Nobody knows how many jobs AI will displace or how long the data center buildout will last, and we say so plainly. The point of scenario planning is to find out, before it matters, which of your revenues would hurt most under each outcome, and to decide in advance what you would do. Treasurers do this every year for recessions and hurricanes. The same discipline applies here.

    Why now

    The labor-income exposure is documented at the federal level. A RAND analysis found that 84 percent of federal revenue in 2024 came from individual and payroll taxes, that the loss of high-income white-collar jobs would hit revenue disproportionately, and that higher corporate profits would not offset the decline unless corporate tax rates were roughly doubled. Economists Anton Korinek and Lee Lockwood warn that even modest labor displacement could significantly strain public finances. Local governments sit downstream. One opinion writer argues that states and localities will bear the initial cost of retraining and income support when workers are displaced.

    The boom itself can be a concentration risk. California’s Legislative Analyst found that stock-option withholding from a handful of large tech companies made up about 10 percent of all state income tax withholding in 2025, and warned that withholding could fall if the AI market deteriorates. At the county level, Loudoun County, Virginia, reports that data centers generated 38 percent of its general fund revenue while occupying about 4 percent of its commercial land. Prince William County projected data center revenue of about $550 million in fiscal 2027, roughly 28 percent of general fund tax revenue. Loudoun staff have warned that if the personal property tax rate is not lowered, data centers could exceed 60 percent of the general fund within five to seven years.

    The lesson is not that data centers are bad revenue. It is that a budget resting on one industry needs a plan for what happens if that industry slows down, and a budget resting on wages needs a plan for what happens if wages shrink.

    The worksheet

    1. Map your revenue by source and by exposure. List every major revenue line: property tax, sales tax, income or payroll taxes where you have them, business taxes, fees and intergovernmental aid. For each, note how much depends on (a) wages in office and professional jobs in your area, (b) a single employer or industry, and (c) data centers or their equipment.
    2. Define three or four scenarios. For example: a baseline, a gradual-displacement scenario, a rapid-displacement scenario, and an investment-bust scenario in which data center equipment values fall sharply. New York City’s Comptroller used five scenarios; a small county can use fewer. Choose ranges your analysts can defend. For illustration only, you might test declines of 10, 25 and 50 percent.
    3. Run each scenario through each revenue line. A wage decline changes income and sales tax receipts. A bust changes personal property tax on equipment. Record the dollar effect and the share of the general fund.
    4. Estimate the demand side. Displaced workers use services. Note which of your programs would see more demand, such as workforce programs, public assistance, and emergency services, and what each would cost.
    5. Set a reserve target. The NYC Comptroller recommended building a reserve to 16 percent of annual tax revenue. Loudoun built a $119.7 million revenue stabilization fund specifically to absorb differences between budgeted and actual data center revenue, and reports it is now fully funded. Your target will differ, but it should be written down.
    6. Set triggers and publish. Decide in advance what happens at each threshold: when to freeze hiring, when to review the tax rate, when to draw on reserves. Then publish the scenarios and triggers each year with the budget. Residents should see the plan before they need it.

    If you host data centers

    Host jurisdictions have an extra set of questions:

    • How much of the general fund comes from data centers, and what is the trend? Track the share every year, not just the dollar amount.
    • What happens if equipment is replaced less often, or values fall? Equipment, not buildings, drives much of the revenue. The Economist notes that more than half of AI spending has gone to servers and chips that become obsolete within a few years.
    • Are you spending windfalls on permanent commitments? Treat unexpected revenue as one-time money unless you can justify it as durable. Loudoun staff have framed lowering the tax rate on data center equipment as the strongest tool for managing dependence.
    • Is relief visible to residents? Loudoun has cut its real estate tax rate substantially over the past decade, a real benefit residents can see.

    What not to do

    • Do not turn scenarios into predictions. A scenario is a stress test. Label it that way.
    • Do not wait for certainty. The range of forecasts is wide, and it will stay wide.
    • Do not model only the downside. Include a scenario in which AI strengthens your local economy, so the worksheet is credible to everyone.

    The case against our case

    • Holding reserves costs services today. Every dollar in reserve is a dollar not spent on schools or roads now. Reasonable officials will weigh the insurance value against current needs, and the right level is a local judgment.
    • The effect may look like earlier technology shifts. Economists David Agrawal and William Fox argue that AI reinforces rather than overturns the classic tradeoffs in state and local public finance, much as e-commerce and remote work did. If so, familiar tools may be enough.
    • Data center revenue is real and useful. Industry groups argue that without it, residents would pay far more. NetChoice, a trade association, says the alternative for Loudoun homeowners is a tax bill nearly double today’s. That argument deserves weight, and it is exactly why the downside needs a plan.
    • Forecasts are uncertain. The NYC report itself notes how hard it is to tie current trends to AI with certainty. We agree, and that is why the worksheet uses ranges.

    The role no one holds

    In most governments, nobody is assigned to ask these questions across revenue, workforce and infrastructure at once. Treasurers see revenue. Economic development officers see ribbon-cuttings. Workforce boards see training. A technology leader’s job is to sit across all three. If you are a treasurer, you do not need to wait for that role to exist: start with step one this quarter. We are glad to share a blank version of the worksheet. Our proposal is at department.technology/.