The debate over artificial intelligence has become divided between two powerful ideas: that we need a “moratorium” to slow AI until we can make it safe, and that regulating AI will slow the innovation we need for our economy and national security. Both concerns are legitimate. But “moratorium” and “slowing AI growth” are divisive terms: they reassure those concerned about AI's safety, economic and environmental consequences while alarming those who see AI competition and national security as urgent priorities. We need a different starting point—meaningful regulation that provides the safeguards necessary to protect people, communities and the environment while creating the conditions for AI to continue growing. The experiences of Europe and China challenge the assumption that regulation must slow AI.
China has combined increasingly detailed, technology-specific AI regulation with substantial government support for AI development. Recent studies provide evidence that this approach can promote innovation. One 2026 study of Chinese companies from 2017 through 2023 found that the creation of national AI innovation and development pilot zones significantly increased corporate digital transformation, in part through increased research and development, human capital and reduced transition costs. Another 2026 study, covering Chinese manufacturing companies from 2014 through 2024, found that the same policy significantly promoted the adoption of intelligent technologies.
Europe provides a different example. The European Union has adopted a risk-based AI regulatory system while AI investment continues to expand. AI accounted for 27% of European venture-capital investment in 2025, according to a European Commission report. More than 4,100 European AI startups in ten strategic industrial sectors had a combined enterprise value of €161 billion in 2025, according to a second European Commission report.
None of this proves that regulation automatically accelerates AI. Europe still faces significant challenges in financing and scaling AI companies, and regulation is only one factor affecting innovation. But these experiences undermine the assumption that regulation and AI growth are inherently incompatible.
The objective should be to regulate AI in ways that make it safer, fairer and more sustainable while allowing it to grow faster.
The scale of the challenge is already enormous. The International Energy Agency reports that electricity consumption by AI-focused data centers increased about 50% in 2025. Overall data-center electricity consumption is projected to nearly double, from about 485 terawatt-hours in 2025 to 950 TWh in 2030.
That growth makes regulation of AI's physical infrastructure increasingly important. AI requires enormous amounts of electricity, as well as land, water, transmission capacity and other resources. Regulation therefore needs to address environmental impacts and the costs imposed on communities.
But we should not respond by simply putting the brakes on AI.
The benefits of AI could be enormous. The challenge is to make sure those benefits are broadly shared while creating incentives for continued innovation. Regulation should address problems that markets alone may not adequately solve: the safety of increasingly capable AI, the distribution of its economic benefits, effects on workers and communities, and environmental impacts. At the same time, regulation should give developers clear rules, encourage competition and create public confidence in AI.
The current debate in Congress illustrates the difficulty.
Sen. Bernie Sanders and Rep. Alexandria Ocasio-Cortez have proposed the AI Data Center Moratorium Act, which would impose a moratorium on new data centers until Congress establishes safeguards addressing AI safety, workers, electricity costs, environmental impacts and community interests. The proposal reflects legitimate concerns about the speed and consequences of AI development.
But a moratorium based primarily on data-center construction is a blunt way to regulate a technology whose risks can vary enormously.
A huge data center can be running relatively ordinary AI. A much smaller facility could be developing an AI system capable of sophisticated autonomous action.
Infrastructure size is not the same thing as AI risk.
A better approach would classify AI according to the risks created by what the system can actually do.
At the basic level are ordinary AI systems such as today's widely used chatbots and productivity tools. They can be enormously useful but generally operate under substantial human direction.
A second category would be more capable frontier AI systems with substantially greater reasoning, coding, scientific or other capabilities and correspondingly greater potential risks.
The third category is autonomous frontier AI: highly capable systems that can independently plan and act, interact with external systems and pursue objectives with limited human direction. This is where the question of meaningful human control becomes especially important.
The regulatory requirements should rise with the risk.
Some applications may need to be prohibited altogether. Others may require extensive safety measures during both development and deployment. Some could be developed with relatively few restrictions but require rigorous testing and safeguards before deployment. Low-risk AI could remain primarily subject to ordinary safeguards.
This is broadly consistent with the logic behind Europe's AI Act, which uses four levels of risk rather than treating every AI application alike. It also provides regulatory sandboxes and testing environments intended to allow innovation while addressing safety requirements.
The highest-risk systems deserve particular attention.
Sen. Sanders and Rep. Greg Casar have proposed the Ban Artificial Superintelligence Act, which would permanently prohibit the development and deployment of what they call artificial superintelligence and temporarily pause advanced AI development until a federal regulator establishes safety rules. The proposal also calls for international agreements to prevent superintelligence from being developed elsewhere.
The concern about losing human control deserves serious attention. But intelligence by itself is not necessarily the danger. A machine that is vastly more capable than humans could be extraordinarily beneficial if humans retain effective control over what it can do.
The danger begins when increasingly capable machines are given the autonomy, authority and access to act beyond meaningful human control.
That principle should have a particularly clear application to lethal force. AI systems should not have independent authority to select human targets or to make and execute decisions to apply lethal force against people. Humans must retain meaningful judgment and effective control over any system capable of applying lethal force. The United States should also work with other nations toward international agreements establishing the same principle.
This does not mean that AI could never be used in military or other dangerous environments. AI can assist humans in detecting threats, analyzing information and defending against attacks. The critical distinction is between AI assistance and AI authority.
Getting regulation right will not be easy. Rules that are too weak will fail to protect people. Rules that are too broad or inflexible could prevent beneficial applications, reduce competition and push development elsewhere.
That is why the debate needs to move beyond the language of “moratoriums” and “slowing AI.” Those words frame the choice as a conflict between safety and innovation, when that should not be the choice. A moratorium may be an appropriate response to an immediate and specific danger, but making “slowing AI” the objective risks putting those concerned about safety on one side and those who see AI as essential to innovation and national security on the other. We need to focus instead on how to meaningfully regulate AI—creating the safeguards necessary to protect people, communities and the environment while providing the conditions for AI to continue growing.
The question is: “How do we regulate AI so that safety, fairness and environmental responsibility reinforce rather than undermine innovation?”
Europe and China offer different experiences, not definitive answers. Neither demonstrates that regulation automatically makes AI grow faster. But both challenge the assumption that regulation must slow it down.
The United States should learn from those experiences and build a regulatory system designed not to stop AI's growth, but to make that growth safer, broader and more sustainable. This approach has the potential to bring together those concerned about AI's safety, economic and environmental consequences with those focused on innovation and growth. Finding that common ground is essential to achieving any of these goals.
Michael A. Soderstrand is a retired professor of Electrical and Computer Engineering and currently lives in Columbus, Ohio.
