The AI Wall Has a Physical Blind Spot
Export controls can restrict chips and models, but not the energy, water, and land the AI race actually runs on — and in that material world, American democracy is the constraint China doesn't carry.
In June, China lowered a data center into the sea off Shanghai. It runs on offshore wind, cools itself with seawater, and uses no freshwater at all. No community meeting was held, no zoning board voted, no utility commission pushed back. It is now operating [1]. In the United States over roughly the same period, communities blocked or delayed more than $100 billion in AI data center projects across at least 11 states [2][3]. New York enacted the first statewide moratorium on certain AI data centers. California, Maine, and South Carolina are pursuing their own bans [4][5]. Idaho legislators mandated that AI data centers provide their own power sources after Idaho Power raised rates 7.48% to cover grid strain from a single Meta facility [6]. PJM Interconnection, which serves 67 million people across 13 states, proposed emergency power cuts to data centers — and the tech firms rejected the proposal as discriminatory [7]. The two images sit on opposite sides of the same strategic competition. The U.S. government has spent years building a wall around AI — export controls on chips, restrictions on model access, a thickening set of legal and regulatory barriers meant to keep advanced capability out of China's hands. That wall has a blind spot, and the blind spot is physical. The U.S.-China Economic and Security Review Commission, the government's own advisory body, put the problem plainly earlier this year: export controls target the "digital loop" of chips and models but are "not well suited to addressing the physical loop of deployment-driven data creation" [8]. The physical loop is energy, water, land — the material substrate on which the entire AI enterprise runs. And in that loop, the containment strategy is colliding with something the digital wall never had to contend with: democratic accountability. The numbers are stark. American opposition to nearby data center construction rose from 28% to 41% in early 2026, with 53% now opposing a data center in their own neighborhood — a shift the polling firm Public First described as unusually rapid [5]. Microsoft listed "hyper-local dissent" as an operational risk in its securities filings [2]. Bernie Sanders called for a national moratorium. This is not a fringe movement; it is a broad, accelerating rebellion against the material footprint of the AI buildout. The buildout's defenders are well-funded and organized. The Manhattan Institute has been actively countering community opposition, with senior fellow Judge Glock arguing that data center resistance rests on incorrect assumptions and that the facilities have negligible effect on local electricity prices [3]. Amazon commissioned research from Energy and Environmental Economics, Inc. to argue that utility rate hikes are mitigable [9]. The argument, repeated by industry and its allies in Washington, is that America must build at all costs to win the AI race, and any regulation that slows the buildout is a national security risk [10]. That last move — the pivot from commercial dispute to national security imperative — is where the contradiction sharpens. Data center companies are now seeking EPA environmental rule exemptions by explicitly citing "national security" and "AI supremacy." Novva's CEO admitted the tactic openly.
Almost everybody would claim it’s some kind of national security issue. — Wes Swenson
The same national-security urgency that justifies the chip wall is now the lever being used to bypass environmental rules in the physical world. The framing that says America must contain China's AI access is the same framing that says America must override local objections to the data centers that containment requires. The asymmetry with China is total. Beijing sites data centers with subsidized power and no community veto [11]. It is leveraging renewable energy explicitly to lower AI computation costs, with Roland Berger's Denis Depoux noting China can lower AI computation costs substantially by pairing green energy with data centers — and is selling that model to the Global South as a cooperative alternative to U.S. protectionism [12]. The underwater data center off Shanghai is not a one-off; it is a template for infrastructure siting that faces no democratic friction. Meanwhile, the digital wall itself is porous in ways that make the physical rebellion more consequential. Chinese open-source models overtook American ones in global downloads by late 2025, with Alibaba's Qwen alone capturing over half of worldwide open-source downloads by mid-2026 [13][14]. The PLA has been using model distillation — extracting capabilities from GPT-3.5 and Claude 3 Haiku — to build military systems for cyber warfare and drone navigation [15]. China's chip self-sufficiency rose from 10% in 2021 to 41% in 2026, projected to reach 86% by 2030 — a trajectory Huawei's Xu Zhijun attributed directly to U.S. pressure [16].
If the US hadn't forced our country, our companies, and our industry, we wouldn't have done something like this. — Xu Zhijun
None of this means the digital wall is pointless. It means the wall is only half the battlefield, and the other half is governed by rules the United States cannot unilaterally set. In the physical loop, the constraint on American AI infrastructure is not Chinese competition but American democracy itself — the zoning boards, the utility commissions, the state legislatures, the neighbors who show up at hearings. These are not obstacles to be cleared; they are the mechanism by which communities decide what gets built in their own backyards. They are, in other words, the thing the wall is supposed to protect. The double bind is this: the national-security framing demands an unconditional buildout, and the buildout's neighbors are rising against it. The same urgency is being used to seek environmental-rule exemptions from the democratic process that stands in its way. China faces no such bind. It can lower a data center into the sea and call it done. The United States cannot — and the commission's blind spot is that it never accounted for the possibility that the harder half of the containment strategy would be convincing Americans to host it.
- 1. China Launches World's First Wind-Powered Underwater Data Center
- 2. U.S. Communities Block $98 Billion in AI Data Centers
- 3. Opposition Blocks $100 Billion in AI Data Center Projects
- 4. New York Bans AI Data Centers as Michigan Project Begins
- 5. US Opposition to Local Data Centers Rises Amid Policy Shifts
- 6. Idaho Legislators Target AI Data Centers Amid Energy Crisis
- 7. AI Data Centers Strain US Power Grid and Spark Legislative Action
- 8. US Commission Warns China's Open-Source AI Threatens US Leadership
- 9. US Communities Oppose AI Data Centers Over Resource Concerns
- 10. U.S. Data Centers Expand to Lead Global AI Race
- 11. China Uses Chip Clusters and Cheap Energy to Rival Nvidia
- 12. Roland Berger Director Highlights China's AI and Green Energy Synergy
- 13. Chinese Developers Overtake U.S. in Open-Access AI Market
- 14. Chinese AI Ecosystem Surges with Open-Source Model Dominance
- 15. China Uses U.S. AI Models to Build Defense Systems
- 16. US Closes Export Loophole for AI Chips to China