Anthropic Warns US: Build AI Power Responsibly, Not Like China

The artificial intelligence firm Anthropic has issued a clear warning to the United States: emulating China's rapid, state-driven approach to building and permitting large-scale infrastructure would be a strategic mistake, despite the urgent need to support the burgeoning AI industry. In a recent report, the Amazon-backed startup argued that while the U.S. must accelerate its infrastructure development, particularly for energy, it should not sacrifice its regulatory and environmental standards to do so.[1] This position highlights a central tension in the global AI race: the need for speed versus the importance of responsible and sustainable development. Anthropic's stance underscores the significant infrastructure demands of advanced AI and the geopolitical competition that frames the industry's future.

At the heart of Anthropic's argument is the immense and growing energy requirement for training and deploying frontier AI models. The company projects that the total demand for AI training in the U.S. could reach 20-25 gigawatts by 2028.[2] To meet this, Anthropic has recommended an ambitious national goal of adding 50 gigawatts of dedicated power capacity by 2027.[3][4] This demand for stable and reliable power is a critical bottleneck for American AI leadership.[2] The firm points to a stark contrast with China, which added over 400 gigawatts of power capacity in a single year, compared to the tens of gigawatts added in the U.S.[2][1] This disparity, Anthropic warns, is a cause for concern as China recognizes the geopolitical and economic importance of AI and is building its own infrastructure at a rapid pace.[1]

However, Anthropic explicitly cautions against adopting China's methods to close this gap.[1] The report notes that while both countries have permitting processes for infrastructure projects, China's regulators complete them in a fraction of the time—typically 3 to 6 months.[1] This speed is attributed to a political system that prioritizes projects aligned with state objectives, often leading officials to skirt environmental reviews and shirk enforcement.[1] In contrast, the U.S. process, governed by laws like the National Environmental Policy Act (NEPA), is significantly longer, with a median timeline of 2.2 years for major project environmental impact statements.[1] Anthropic's position is that while the U.S. must streamline its own regulatory challenges that delay energy projects, it should do so without compromising its foundational principles.[2] The company advocates for accelerating permitting, supporting targeted transmission projects, and working with utilities to speed up grid interconnection, all within the existing American framework.[2]

Anthropic’s recommendations extend beyond energy to encompass a broader strategy for securing U.S. leadership in AI. The company has submitted proposals to the White House's Office of Science and Technology Policy focusing on several key pillars.[5][3] These include enhancing national security through robust testing of AI models, strengthening export controls on critical technologies like advanced semiconductors to prevent access by adversaries, and improving security at AI labs through collaboration with intelligence agencies.[5][3][4] These defensive measures are part of a strategy to protect U.S. technological assets while simultaneously investing in a domestic ecosystem that fosters innovation and ensures economic benefits are widely shared.[6] This involves accelerating AI adoption within the government and modernizing economic data collection to prepare for large-scale economic shifts driven by AI.[5][3]

Ultimately, Anthropic's message is one of strategic patience and principled acceleration. The firm believes the United States possesses the economic strength and innovative spirit to meet the infrastructure challenges of the AI era.[2] The proposed path forward involves leveraging existing federal authorities to cut through regulatory red tape without dismantling the system entirely.[2] By making federal lands available for AI infrastructure, expediting environmental reviews for certain projects, and fostering public-private partnerships, the U.S. can build the necessary foundation for AI leadership.[2] This approach seeks to strike a difficult but necessary balance, ensuring that in the race to build the future of artificial intelligence, the United States does not sacrifice the very values and systems it aims to protect. The firm's detailed recommendations underscore a belief that long-term, sustainable leadership in AI is not just about building faster, but building smarter and more responsibly.