The Surrender Doctrine: Why a U.S. Data Center Moratorium Would Hand the Twenty-First Century to Beijing.

On April 24, 2026, Maine Governor Janet Mills vetoed LD 307 — a bill that would have made Maine the first state in U.S. history to impose a statewide moratorium on data centers above 20 megawatts.¹ The veto bought the United States a few months. It did not end the debate. By the time Mills's pen hit paper, more than three hundred data-center-related bills had been filed across thirty states in the first six weeks of the year alone, with at least a dozen state legislatures and sixty-three local governments actively considering pauses on new construction.²³ On April 29, Senator Bernie Sanders introduced the federal "Artificial Intelligence Data Center Moratorium Act," with Representative Alexandria Ocasio-Cortez preparing a House companion.⁴

The arguments offered are not unreasonable on their face. Electricity rates are climbing in Virginia, Georgia, and Texas. Aquifers are stressed. Rural communities are blindsided by 1,000-acre announcements they had no role in approving. These are real problems, and they deserve real policy responses — utility rate-class reform, water disclosure requirements, transparent zoning, fair grid-cost allocation. We have argued for such reforms in this publication before.

But a moratorium is not a reform. A moratorium is a stop-work order issued in the middle of the most consequential technological mobilization since the Manhattan Project — and the editors of AI Newscast believe that issuing one would constitute one of the gravest unforced strategic errors in postwar American history. The thesis of this essay is simple, and we will defend it across the next several thousand words: halting U.S. data center construction would be a gift to the People's Republic of China, a country that has already won the solar manufacturing war, has already won the humanoid robotics war, is rapidly closing the AI compute gap, and has openly declared a Manhattan-Project-scale program to put solar power stations the size of the Three Gorges Dam into geostationary orbit. To pause now is to forfeit the only race that matters.

§ 01 — The Buildout RaceHow the Numbers Actually Look in 2026

Before debating whether to pause, it is worth establishing where the two countries actually stand. The conventional Silicon Valley narrative — that America is comfortably ahead in AI and merely needs to manage growing pains — has not survived contact with 2026 data.

By the close of 2025, China had installed roughly 32 gigawatts of operational data center capacity. By year-end 2026, Rystad Energy projects that figure will climb to 40 GW, and to 60 GW by 2030 — a near doubling in five years, with data center electricity consumption rising to 289 terawatt-hours and growing at a 19 % annual clip.⁵ The 15th Five-Year Plan, which took effect in January, designates data center buildout a national strategic priority and mandates that all facilities in the eight national computing hubs source at least 80 % of their power from renewables.⁵

The United States, by contrast, is stalling. Sightline Climate data cited by Bloomberg shows that of the 12 GW of U.S. data center capacity announced for completion in 2026, only about a third has broken ground; analysts now project a 7-gigawatt shortfall against committed delivery.⁶ A $2 billion campus can sit idle waiting on a $40 million transformer order — and a significant share of those transformers, switchgear, and batteries are manufactured in China.⁷ Hyperscalers — Alphabet, Amazon, Meta, Microsoft — are committing roughly $725 billion of capital expenditure for 2026, a 77 % jump year-over-year.⁸ The bottleneck is not capital. The bottleneck is permission and physical infrastructure.

The picture sharpens further when you look at cost. According to industry reporting, Chinese data centers pay less than half the electricity rates of American facilities, and projects in China move from approval to operation in months rather than years.⁹ By 2030, China is projected to hold roughly 400 gigawatts of spare power capacity — triple the global expected demand of the entire data center fleet.⁹ American grid operators, by contrast, are warning of a 44 GW shortfall over the next three years.⁹

This is the situation a moratorium would freeze in place. Proponents argue a pause is short. But construction pipelines are measured in years; equipment supply chains in decades; talent migration in careers. A two-year halt on permits is not a two-year delay — it is a structural surrender of a buildout cycle that determines which country trains the largest models from 2027 through 2032, the years most experts expect AGI-class systems to arrive.

§ 02 — The Solar War, Already LostWhat "Just Building Solar" Looks Like When You Don't Make Solar

The most popular counterargument to massive U.S. data center buildout is that we should slow down because the grid cannot handle it cleanly. The implicit assumption is that energy is a domestic question. It is not. Energy is now a manufacturing question, and the manufacturing belongs to China.

According to the International Energy Agency, China's share of every manufacturing stage of solar panels — polysilicon, ingots, wafers, cells, modules — exceeds 80 %, with wafer and ingot production approaching 95 %.¹⁰ One out of every seven solar panels produced anywhere on Earth comes from a single Chinese facility.¹⁰ Beijing has invested over $130 billion in the sector in recent years; the top twenty-five global manufacturers shipped a combined 687 GW of modules in 2024, and Chinese giants Jinko Solar and LONGi tied for first place in 2025 with 80–90 GW each.¹¹¹² Chinese solar cell exports grew 73 % in 2025 alone.¹³

This is not a matter of catching up. Wood Mackenzie and the IEA both project China will hold over 80 % of global manufacturing capacity through at least 2026, and Chinese capacity (1,200 GW in 2025) is already nearly double total global installation demand of 650 GW.¹⁴ The competitive gap is widening, not narrowing.

The implication is direct and uncomfortable. Every gigawatt of solar that the United States, the European Union, India, or anyone else installs to power AI data centers is — for the foreseeable future — a gigawatt that flows through Chinese factories. A moratorium that delays U.S. data center construction does not redirect demand to American solar manufacturing. It simply removes American demand from the market entirely. The Chinese factories continue to run; they just sell their panels into Africa, the Gulf, Latin America, and Southeast Asia, building deeper energy diplomacy in every region. Meanwhile, U.S. renewable installers — many of them small businesses in red states — lose project pipelines.

China's solar manufacturing expansion has been driven by high margins for polysilicon, technology upgrades and policy support… China will still dominate the global solar supply chain and continue to widen the technology and cost gap with competitors. — Huaiyan Sun, senior consultant, Wood Mackenzie¹⁴

§ 03 — The Robotics War, Also LostEmbodied AI and the Manufacturing Floor of 2030

If the solar story is about the energy that powers AI, the robotics story is about the bodies that AI will inhabit. Here the picture is, if anything, more lopsided.

According to Omdia and Rest of World reporting, Chinese companies accounted for roughly 90 % of all humanoid robots sold worldwide in 2025, with six of the top-selling firms based in China.¹⁵ Shanghai-based AgiBot shipped about 5,168 units; Hangzhou's Unitree Robotics shipped over 5,500 — each individually exceeding Tesla's stated Optimus production target for the year, which Tesla did not meet.¹⁵¹⁶ Unitree launched its R1 humanoid in July 2025 at $5,900, a price that was considered impossible by Western analysts a year prior.¹⁷

China recorded over 610 robotics investment deals in just the first nine months of 2025, totaling roughly $7 billion — a 250 % year-over-year increase.¹⁷ The International Federation of Robotics reports China commanded 54 % of global industrial robot deployments in the same period.¹⁸ Tesla CEO Elon Musk himself acknowledged at the World Economic Forum: "China is very good at AI, very good at manufacturing… To the best of our knowledge, we don't see any significant competitors outside of China."¹⁶

Morgan Stanley estimates the humanoid robot market will reach $5 trillion by 2050, with more than one billion units in use.¹⁹ The bank's Sheng Zhong puts it bluntly: "It is becoming apparent that national support for 'embodied AI' may be far greater in China than in any other nation, driving continued innovation and capital formation. China's lead in AI-robotics may need to widen before rivals, including the U.S., pay closer attention."¹⁹

Why does this matter for a data-center moratorium discussion? Because humanoid robots — the next great industrial platform — are trained on data centers. Embodied AI requires colossal compute for simulation, reinforcement learning, and continuous fleet learning. China's lead in physical robot production combined with its accelerating compute buildout means the country is on track to control both the brains and the bodies of the next industrial revolution. Pausing American compute does not slow that flywheel. It accelerates it relatively.

§ 04 — The Orbital FrontierWhy a Moratorium Forfeits the Final High Ground

Here is the part of the story that almost no participant in the moratorium debate has internalized: the next layer of the AI race is not on the ground. It is in orbit. And both the United States and China are now openly racing to build it — except that one of them has explicitly framed the program as a Manhattan-Project-scale national priority.

In late 2024, Long Lehao, a senior rocket scientist and member of the Chinese Academy of Engineering, publicly described China's Zhuri ("sun-chasing") project as "another Three Gorges Dam project above the Earth" — a 1-kilometer-wide solar array in geostationary orbit, with a megawatt-class in-orbit demonstration targeted for around 2030.²⁰²¹ Long compared the program directly to the Manhattan Project. The Three Gorges Dam, for context, is the largest hydropower facility on Earth, generating roughly 100 billion kilowatt-hours per year. Long's argument is that a single orbital array, 1 km wide along the geostationary belt, could capture annual energy "equivalent to the total amount of oil that can be extracted from the Earth."²⁰

Meanwhile, the American private sector has begun moving — but without national-program coordination. In November 2025, Google announced Project Suncatcher, planning solar-powered satellite constellations carrying its TPU-class AI chips, with a demonstration mission planned for 2027.²² Around the same time, the startup Starcloud launched a 60-kilogram satellite carrying an NVIDIA H100, becoming the first company to train an LLM in space; in February 2026, Starcloud filed with the FCC for a constellation of up to 88,000 satellites.²³ SpaceX has filed for up to one million orbital data center satellites; Blue Origin filed for 51,600.²⁴ On April 27, 2026, Meta signed a landmark agreement with Overview Energy to reserve up to 1 GW of space-based solar power, beamed via near-infrared lasers to terrestrial solar farms — the first major hyperscaler space-power procurement contract in history.²⁵

The orbital frontier matters because it is the only domain where the constraints of terrestrial NIMBYism, transformer shortages, water tables, and aquifer depletion do not apply. Solar irradiance in low Earth orbit is roughly 36 % higher than on the surface; in geostationary orbit, sunlight is essentially continuous.²³ A reasonably sized array can produce continuous gigawatt-class power. There is no permitting fight, no rural Michigan rally, no veto.

But getting to orbit cheaply requires industrial momentum on the ground — chip fabs, launch infrastructure, AI software stacks, and yes, terrestrial data centers that can pay for the R&D and serve as the demand backbone for orbital expansion. A moratorium would freeze that momentum at exactly the moment it most needs to compound. If American hyperscalers cannot build domestically, capital flows to allied jurisdictions or, more darkly, withdraws from the AI buildout altogether. Project Suncatcher, Meta's Overview deal, Starcloud, Blue Origin's filings — these are downstream of robust terrestrial compute economics. Strangle the terrestrial buildout and you starve the orbital one.

§ 05 — The Manhattan Project FrameWhat That Comparison Actually Means, and What It Doesn't

This publication does not invoke the Manhattan Project lightly. The original program took 130,000 workers, $2 billion in 1940s dollars, and three years to produce the weapons that ended the war in the Pacific. It was the most concentrated act of technological mobilization in U.S. history. The comparison to AI has been made — most prominently — by serious figures. The bipartisan U.S.-China Economic and Security Review Commission's 2024 annual report to Congress recommended, as its top item, that "Congress establish and fund a Manhattan Project–like program dedicated to racing to and acquiring an Artificial General Intelligence (AGI) capability."²⁶ Outgoing National Security Advisor Jake Sullivan, days before the Trump inauguration, told Axios that "staying ahead in the AI arms race makes the Manhattan Project during World War II seem tiny."²⁷

We want to be careful about two things. First, the analogy is imperfect: the original Manhattan Project was a weapons program with a clear, single deliverable. AGI is more like electricity — a general-purpose technology. Second, serious analysts at RAND, AI Frontiers, and the Foundation for American Innovation have raised legitimate concerns about whether wartime-style framing actually increases escalation risk with China.²⁸²⁹ RAND's Matt Chessen has argued that an "Apollo-style moonshot" framing may be safer than a Manhattan one — peaceful, scientific, civilian-led — and we agree with much of that critique on framing.²⁸

But the underlying point survives: the United States is in a generational technological competition whose outcome will be determined by industrial capacity, compute infrastructure, energy production, and engineering talent. That is exactly the dimension in which moratoriums bite. You cannot field a Manhattan Project — or an Apollo program, or any equivalent — while simultaneously banning the construction of the facilities that produce its core inputs. The two postures are mutually exclusive.

§ 06 — A Better PathWhat the U.S. Should Do Instead

A serious policy response would start by separating three problems that moratorium bills tend to conflate: local impact, cost-shifting, and strategic capacity. Each deserves a distinct tool.

Local impact: real, addressable, not a national-security argument.

Aquifer depletion, opaque NDAs, and absent rural permitting frameworks are state and local issues that deserve state and local solutions. South Dakota's SB 135 and Washington's SB 5982 — neither of which is a moratorium — show the model: mandatory disclosure of water and energy use, dedicated rate classes for hyperscalers, and prohibition on NDAs between data center operators and local government water/electricity contracts. New Jersey's A796/S731 requires data centers using 100+ MW to pay at least 85 % of their service costs under a 10-year commitment.² These are surgical, effective, and compatible with a pro-buildout posture.

Cost-shifting: solvable through ratepayer protection.

The fear that residents will subsidize hyperscalers' grid upgrades is legitimate and avoidable. The principle should be: the load that drives the upgrade pays for the upgrade. Eighteen states have introduced bills creating special rate classes for large energy users in 2026.² This is the right reform. None of it requires halting construction.

Strategic capacity: actively build, federally coordinate.

Whether one calls it a Manhattan Project, an Apollo Program, or an Eisenhower-style highways act, the United States needs federal coordination on three dimensions: a permitting fast-lane for grid-interconnected, water-disclosed, ratepayer-neutral data centers; a domestic transformer and switchgear manufacturing rebuild (currently a chokepoint inflicted by China); and a coordinated push on next-generation power — Small Modular Reactors, geothermal at scale, and orbital solar. Meta's Overview deal, Microsoft's nuclear agreements, and Google's Suncatcher show the private sector is ready. What's missing is a federal program that matches China's industrial-policy coherence.

§ 07 — Closing ArgumentWhat Pausing Actually Means

In the spring of 1940, Vannevar Bush walked into the Oval Office with a single-page memo proposing what would become the Office of Scientific Research and Development. Roosevelt approved it in ten minutes. The country was not yet at war. Most Americans considered the European conflict distant and someone else's problem. Bush's argument was that by the time the threat was unambiguous, the lead time to respond would already be lost. He was right.

The 2026 debate over a U.S. data center moratorium is, in its strategic shape, a debate about the same lead time. The People's Republic of China is not pausing. It is doubling its compute capacity, holding 80 %-plus of the solar supply chain, controlling 90 % of humanoid robotics, planning a megawatt-class space solar demonstration, and announcing all of this as deliberate national policy under its 15th Five-Year Plan. The Chinese government and its private firms, as the Atlantic Council's Frederick Kempe noted, "work hand in glove."²⁷ That coordination is the structural advantage the United States must counter — not by copying authoritarian methods, but by getting out of its own way.

A moratorium does not address the legitimate grievances of residents in Loudoun County, Memphis, Abilene, or rural Maine. It does not lower their electricity bills, restore their aquifers, or give them the public hearings they should have had two years ago. What it does is hand the next decade of artificial intelligence — the technology that will decide whether the twenty-first century is shaped by liberal democracy or by Beijing's model of integrated state capitalism — to a competitor that has already mapped the next two terrains.

The United States has done this before. In the late 1940s, faced with a Soviet adversary that had mobilized its entire society around a technological challenge, the country did not pause. It built. It built the National Science Foundation, the Atomic Energy Commission, the Interstate Highway System, NASA, ARPANET. Every one of those projects had local critics. Every one produced disruption. None were paused. All were, in retrospect, the reason the United States entered the twenty-first century as the world's leading economy and democracy.

The choice in 2026 is the same choice. The data centers will be built. The only question is whether they are built in Texas and Ohio, or in Inner Mongolia and Hangzhou. There is no third option in which neither side builds and we all live in pastoral equilibrium. There is only the option where America builds, and the option where America surrenders.

This publication believes the choice is obvious. Reform, yes. Disclosure, yes. Ratepayer protection, yes. Permitting reform, yes. But pause? Pause is the surrender doctrine. And the United States does not have the luxury, in the spring of 2026, of surrendering to anything.