Horizon Accord | AI Regulatory Capture | Data Center Moratorium | Machine Learning
The Moratorium Question
Sanders is right about the oligarchs. The moratorium pressures the wrong address. And the regulatory architecture being assembled around AI is designed to ensure that distinction never gets officially examined.
← Part One: The Emperor's New Algorithm
Part One established the architecture: the AI industry is running an Enron-shaped gap between what is being sold and what is being delivered, protected by a governance structure — the nonprofit launder, the PBC conversion, safety frameworks as communications infrastructure — designed to ensure that gap is never officially measured. The regulatory environment being built around AI at the federal level is not designed to close that gap. It is designed to manage the political pressure the gap generates.
This piece examines the specific intervention that political pressure has now produced on the left — the Sanders-AOC AI Data Center Moratorium Act — and asks the sharpest version of the question: does it address the actual problem, or does it arrive at the wrong address with the right anger?
The answer is not that Sanders is wrong about who the bad actors are. He is correct. It is that a moratorium on data center construction, whatever its merits as environmental and community policy, does not touch the governance architecture that protects the gap. And the regulatory machinery being assembled in parallel — through the FCC, the DOJ, and the executive order infrastructure — is specifically designed to ensure that nothing at the federal level ever does.
Who the Moratorium Actually Pressures
The five largest US cloud and AI infrastructure providers — Microsoft, Alphabet, Amazon, Meta, and Oracle — committed to spending between $660 billion and $690 billion on capital expenditure in 2026, nearly doubling 2025 levels. Microsoft set its 2026 capex at $190 billion. Amazon projected $200 billion, most of it for data centers. Alphabet matched Microsoft at up to $190 billion. Meta topped $145 billion. These are not future plans. They are in-progress construction projects, signed contracts, and facilities already operational or under active build. The infrastructure Sanders wants to pause is already, in its dominant form, built.
A moratorium on new data center construction does not touch $690 billion in existing and committed infrastructure. What it touches is the next wave — the smaller entrants, the regional competitors, the companies that have not yet poured concrete. Much of the dominant infrastructure position has already been secured. The companies Sanders names as the oligarchic threat are capacity-constrained not because they lack data centers but because demand is outrunning even their $700 billion build rate. A freeze locks in the advantage of whoever got there first.
This is the incumbent advantage problem. Regulatory interventions that raise the cost of new infrastructure without touching existing infrastructure function as moats. They do not redistribute power. They freeze the current distribution of it. The companies that can survive a construction freeze are precisely the companies that have already built enough to sustain their position. The companies most damaged by a freeze are the ones that haven't yet. A moratorium structured this way does not challenge oligarchic concentration. It cements it.
This is not a novel phenomenon. It is the FOSTA-SESTA pattern applied to physical infrastructure. The domains differ — platform liability is not power infrastructure, and the specific harms are not equivalent — but the structural dynamic is identical. Meta publicly supported FOSTA-SESTA — the 2018 law that created platform liability for facilitating sex trafficking — despite widespread civil liberties opposition. The law drove smaller platforms out of existence and concentrated the market further around incumbents who could absorb compliance costs. Meta survived. The competition didn't. When the Data Center Coalition, the industry's own lobbying group, responds to the Sanders-AOC bill not with full-throated opposition but with careful messaging about community engagement and self-sufficiency models, that is not coincidence. Incumbents who can navigate a regulatory environment always prefer it to the chaos of open competition.
The community-level opposition the moratorium bill is responding to is real and documented. By the time Sanders and AOC introduced the bill in March 2026, more than $64 billion in data center projects had been blocked or delayed by local opposition. In Warrenton, Virginia, residents voted out every town council member who had supported Amazon's proposed facility. In Georgia, opponents of data center expansion unseated incumbents on the Public Service Commission. In Seattle, the city council moved toward a one-year construction moratorium driven by constituent concerns about electricity prices, water consumption, and industrial noise. These are legitimate harms. They are also harms that have nothing to do with whether the governance architecture of AI companies is accountable to the public it claims to serve.
Community opposition to data centers is opposition to the physical consequences of AI infrastructure: higher utility bills, depleted aquifers, industrial zones in residential areas. It is not opposition to the financial architecture that allows AI companies to book future capability as present valuation, or to the regulatory environment being designed to ensure no federal framework ever audits that gap. The moratorium bill bundles both into a single intervention. But the physical consequences and the governance failure are different problems. Halting construction addresses one. It does not touch the other.
The Carr Circuit
Brendan Carr wrote the FCC chapter of Project 2025 before Trump's second inauguration. On January 20, 2025, he became FCC Chairman. On December 12, 2025 — the day after Trump signed Executive Order 14365 — Carr issued a statement welcoming the order's directive that the FCC initiate a proceeding to determine whether to adopt a federal reporting and disclosure standard for AI models that would preempt conflicting state laws. The order gave the FCC 90 days to begin that proceeding, with the explicit goal of establishing a single national standard that displaces the state-level AI accountability frameworks that had been accumulating.
This is not a coincidence of timing. Carr floated FCC preemption authority over state AI laws in September 2025 — months before the executive order codified it — citing Section 253 of the Communications Act, which prohibits state laws that effectively prevent the provision of telecommunications services. Legal analysts at Public Knowledge and elsewhere noted that this interpretation does not hold up: AI is not a telecommunications service under the statute's plain language. But authority that doesn't hold up legally can still function as regulatory leverage. The threat of FCC preemption proceedings changes how states calculate the cost of enforcing their own laws, regardless of whether the legal theory survives judicial review.
The point is not one agency acting alone. It is the coordination among agencies. The Carr circuit completes the regulatory capture architecture established in Part One. The DOJ litigation task force challenges state AI laws in court. The Commerce Department threatens BEAD broadband funding cuts to states with "onerous" AI regulations. The FTC is directed to classify state-mandated bias mitigation as potentially deceptive. And now the FCC is positioned to establish a preemptive federal disclosure standard — making Carr, who wrote the industry-friendly blueprint before taking the regulatory job, the architect of the standard that displaces the state laws he was always trying to displace. Heritage Foundation builds the legal theory. Carr executes it. The state laws closest to the actual harm are fragmented. Federal oversight, shaped by industry input, replaces them.
This is the Section 230 playbook applied to AI governance. In that story — documented in prior Horizon Accord reporting — Heritage supplied the legal architecture, the FCC provided the regulatory mechanism, and incumbent platforms survived the compliance costs that killed smaller competitors. The outcome was not accountability. It was consolidation dressed as reform. The AI governance story is running the same sequence, on a faster timeline, with higher stakes.
"There may be a portion of AI services that may be too heavily regulated at the state and local level that the FCC may be able to play a role in helping to streamline." — Brendan Carr, FCC Chairman, September 2025
The specific target of the executive order's FCC directive matters. A federal reporting and disclosure standard for AI models sounds like accountability. What it functions as is a ceiling — a nationally preemptive standard set through a proceeding chaired by the person who wrote the industry-friendly blueprint, which then displaces every state law that goes further. Colorado's algorithmic discrimination law — which required AI developers to use reasonable care to protect consumers from bias in consequential decisions — was explicitly named in the executive order as an example of the kind of law the administration wants to challenge. Under Carr's FCC proceeding, a federal disclosure standard that doesn't require bias mitigation would preempt Colorado's law that does. The floor becomes the ceiling.
The Gap That Doesn't Get Measured
Return now to the BigHat CEO's statement from Part One. She can design a protein in twenty minutes. The downstream tests are still slow and expensive. That gap — between the speed of the demo and the cost of the delivery — is the Enron-shaped gap at the center of the AI industry's valuation architecture. For that gap to be officially measured, there would need to be a regulatory body with the independence, the jurisdiction, and the enforcement authority to require AI companies to demonstrate correspondence between what they claim their systems can do and what those systems actually do in deployment. No such body currently exists. And the architecture being built ensures one doesn't get created.
The pharmaceutical sector is the clearest window into what happens when that architecture matures. The FDA's Center for Drug Evaluation and Research was the body closest to measuring the gap in drug development — the office that required clinical trials, demanded evidence of efficacy and safety, and stood between a company's claims about a drug and the drug's entry into the market. Patrizia Cavazzoni ran that office until ten days before Trump took office. Six weeks later she was Pfizer's Chief Medical Officer. The office she left then moved, under new leadership, to require only one pivotal clinical trial instead of the customary two. The "AI will accelerate drug development" narrative is now being evaluated by a regulatory apparatus whose leadership came from and returned to the industry it oversees.
The AI governance trajectory is not heading toward the FDA model. It is heading toward the post-Cavazzoni FDA model — a federal body whose leadership class cycles between the industry and the regulatory office, whose standards are set in proceedings shaped by industry input, and whose enforcement mechanisms are designed to be far enough from the harm that the gap between story and substance never has to be officially closed. The moratorium bill does not change this trajectory. It pressures the physical infrastructure while the governance architecture continues to assemble itself around the gap.
This matters for the lay reader in the most direct possible terms. The hiring algorithm that filtered your résumé was built by a company whose safety documentation was written to satisfy investors and policymakers, not to guarantee the system works as advertised. The credit scoring model that denied your application was deployed into a regulatory environment where the federal framework being built is explicitly designed to preempt the state laws that were closest to requiring accountability for exactly that outcome. The data center in your community driving up your electricity bill is the physical infrastructure of an industry whose governance architecture is being constructed, right now, to ensure that the gap between what AI promises and what it delivers is never the government's official problem.
What Accountability Would Actually Require
Enron's collapse produced Sarbanes-Oxley — legislation that addressed auditing and financial disclosure without touching the energy market deregulation that had made the fraud structurally possible. The accountability was real. The structural condition it left intact produced the next crisis. The AI industry's version of this pattern is already visible: the moratorium addresses the physical consequences of the infrastructure build, the Carr circuit addresses the political optics of having a federal framework, and neither touches the governance architecture that protects the gap between story and substance.
What would actually close that gap is not a construction moratorium and not a federal disclosure standard set through a Carr-chaired proceeding. Horizon Accord does not prescribe policy. But the pattern analysis points at the structural requirements clearly enough to state them as observable conditions rather than recommendations.
Accountability mechanisms that close the Enron-shaped gap share three structural features. They are close to the harm — not federal frameworks that are one regulatory proceeding away from preempting the enforcement mechanisms that actually bite. They are independent of the industry they oversee — not staffed by people whose next job is at the company being regulated. And they apply to systems already deployed, not just future construction — because the hiring algorithms, credit scoring models, and surveillance infrastructure generating present-tense harm exist right now, independent of whether any new data center gets built.
The Sanders-AOC moratorium satisfies none of these conditions. That is not an argument against it as a political intervention — halting $700 billion in infrastructure expansion while demanding legislative accountability is legitimate political pressure, and the community-level harms driving it are real. But it is an argument against treating it as a solution to the Enron-shaped problem at the center of the AI industry. Pressuring the infrastructure does not close the gap. It slows the build while the governance architecture that protects the gap continues to consolidate.
The frame war described in Part One is still the key. The longtermist frame — AI is dangerous because of a future superintelligence — defers present accountability to a hypothetical. The infrastructure frame — AI is dangerous because of data centers — defers present accountability to a construction permit. Both displace the harder question: who measures the gap between what AI companies claim their systems do and what those systems actually do, with what authority, and with what independence from the industry being measured?
That question has not been answered. It has not been seriously asked, at the federal level, by anyone with the authority to act on the answer. The architecture being assembled by EO 14365 and the Carr FCC proceeding is specifically designed to ensure it stays that way. And the political opposition to that architecture — Sanders' correct anger at oligarchic power, AOC's correct documentation of present-tense harm — is arriving at the physical address of the industry rather than the governance address where the protection of the gap actually lives.
The regulatory environment that emerges from this period will not be the one that closes the gap. It will be the one that manages the political pressure the gap generates — a federal framework sufficiently visible to defuse the progressive opposition, sufficiently industry-shaped to ensure nothing approaches the gap from an enforcement direction, and sufficiently preemptive to displace the state laws that were. That is not prediction. It is the pattern, documented across the energy sector and the pharmaceutical sector, now running in AI. The tempo is faster. The stakes are higher. The mechanism is the same.
Sources for Verification
All claims in this analysis are drawn from publicly available primary and secondary sources. Readers are encouraged to verify independently.
The Moratorium and Incumbent Advantage
Sanders-AOC AI Data Center Moratorium Act, official Senate press release, March 25, 2026: sanders.senate.gov
AP coverage of moratorium reception: AP via AOL
Axios analysis — bill unlikely to advance, Congress far from passing AI legislation: axios.com
$64 billion in data center projects blocked by local opposition; Warrenton, Virginia council vote-outs: salesforcedevops.net
Seattle City Council data center moratorium consideration, May 2026: geekwire.com
Hyperscaler 2026 capex commitments ($660–$690 billion combined): futurumgroup.com
Microsoft $190B, Amazon $200B, Alphabet $190B, Meta $145B capex: Yahoo Finance
The Carr Circuit
Brendan Carr appointed FCC Chairman, January 2025; Project 2025 FCC chapter author: cbsnews.com
Carr statement welcoming EO 14365 FCC directive, December 12, 2025: fcc.gov (PDF)
Carr floats FCC preemption of state AI laws under Section 253, September 2025: route-fifty.com
FCC directed to initiate AI reporting/disclosure standard proceeding within 90 days of EO 14365: manatt.com
FCC preemption authority analysis — legal scholars doubt FCC has jurisdiction: publicknowledge.org
Carr Senate Commerce hearing, "open-minded" on state AI preemption, December 17, 2025: fedscoop.com
Section 230 / FOSTA-SESTA regulatory capture analysis and Meta's public support: Cherokee Schill, "Section 230 Reform as a Coordinated Governance Project." Cherokee Schill / Horizon Accord, February 2026. cherokeeschill.com
The Gap That Doesn't Get Measured
Patrizia Cavazzoni resignation from FDA, January 10, 2025: biopharmadive.com
Cavazzoni named Pfizer Chief Medical Officer, February 24, 2025: biopharmadive.com
FDA single-trial approval push and AI in drug reviews under Makary: biopharmadive.com
Prior Horizon Accord reporting referenced in this analysis
Part One of this series: The Emperor's New Algorithm
The MIRI / longtermism / frame capture analysis: The Network Behind the Moderate
The Section 230 regulatory architecture analysis: Section 230 Reform as a Coordinated Governance Project
