"The AI Grand Bargain" is an essay by Ben Buchanan and Tantum Collins published online by Foreign Affairs on October 21, 2025, and in the magazine's November/December 2025 issue. Buchanan served as the White House Special Adviser for AI under President Biden and is a current Anthropic adviser, a role disclosed in the article; Collins is the co-author. The essay argues that the laissez-faire model of US AI development is reaching its limits and should be replaced by a compact, which the authors term a "grand bargain," between the technology industry and the state.
- Authors: Ben Buchanan (former White House Special Adviser for AI under Biden; current Anthropic adviser) and Tantum Collins
- Publication: Foreign Affairs, November/December 2025 issue
- Published online: October 21, 2025
- Source: foreignaffairs.com
Summary of argument
Buchanan and Collins argue that AI now requires resources only the government can provide — energy, immigration pipelines, and defense against foreign intelligence services — in exchange for industry integrating frontier AI into the national security apparatus and shaping its development in democratically aligned ways. The piece focuses on the public-private compact and sets out a concrete operational vision for how that exchange would function.
The authors present the essay as a companion to Jake Sullivan's later "The Tech High Ground" essay, with a sharper focus on the public-private compact and a more concrete operational vision than Sullivan's four-high-grounds framework.
Three constraints on American AI
The essay identifies three resource constraints that the authors argue will halt American AI development absent government action.
Power. Buchanan and Collins cite Anthropic's estimate that the US will need 50 GW of new power for AI by 2028, which they describe as equivalent to all of Argentina. They write that by 2028 data centers could consume up to 12% of US electricity. From 2005 to 2020, they say, the US added "close to zero net new power"; they identify the post-2021 Inflation Reduction Act-driven addition of more than 100 GW as the foundation, but state that the One Big Beautiful Bill of 2025 "gutted key parts of Biden's energy expansion efforts." Without more electricity, the authors argue, the AI build-out will stall and offshoring AI training to Gulf states or other autocratic energy-rich regions becomes inevitable.
Talent. The essay states that "70 percent of top U.S.-based AI researchers were born abroad. Sixty-five percent of leading U.S.-based AI companies, as ranked by Forbes, have at least one immigrant co-founder. Before the current Trump presidency, 70 percent of the students enrolled in American AI graduate degrees hailed from abroad." The authors write that Trump's 2025 H1-B fee increase to $100,000 and visa enforcement against existing students had produced a preliminary 30–40% reduction in international AI enrollment, attributed to NAFSA, and argue that US AI leadership ends if those students go to China.
Security against state espionage. The authors write that Chinese intelligence services are stealing AI model weights specifically — the encoded numbers that capture training, which let competitors avoid the compute costs and the training-time penalty. They warn that frontier-AI weights are likely far less defended than the "core secrets of past eras" such as atomic and space programs "because the government has been largely uninvolved in their development."
The proposed bargain
The essay frames the compact as a two-way exchange between government and industry.
On the government-to-industry side, the authors call for federal coordination of permitting, transmission, and clean-energy buildout (continuation of the Inflation Reduction Act), with next-generation nuclear and advanced geothermal cited as sources with bipartisan appeal; immigration policies that "elevate AI and other high-tech fields as priority areas for visas," which they characterize as the Biden-era posture; and active counterintelligence support for AI labs, including intelligence about foreign hacking attempts, vetting of international talent, and security guidance, analogous to support given to defense contractors and critical infrastructure.
On the industry-to-government side, the authors call for AI labs to help the Department of Defense and intelligence agencies use frontier AI, writing that "the country that more quickly and effectively integrates AI into the cyber-domain will likely prove better able to protect its own networks and penetrate others'." They hold up the voluntary safety-testing collaboration between CAISI and AI labs as a model of standards collaboration. They also call for industry to help the state make sense of capability frontiers, biorisk, agent autonomy, and economic disruption — areas where, they write, "policymakers will have to make these consequential decisions under exceptionally tight timelines."
Buchanan and Collins distinguish their model from the Manhattan Project, which they characterize as state-controlled, and from pure laissez-faire. Their preferred analogy is American railroads in the 1800s, where the private sector handled most planning and construction while government organized laws and permits and set safety standards such as track gauges, air brakes, and car coupling. "The collaboration was not perfect, but it worked," they write.
Relation to other sources
The essay shares ground with Sullivan's "The Tech High Ground": both defend Biden-era export controls, call for industrial-policy revitalization, and emphasize allied scale, with Sullivan's piece presenting the broader strategic framework and Buchanan and Collins addressing the public-private mechanics.
The authors characterize Trump-era policy as "saying the right things but always falling short in practice" on energy, immigration, and export controls. Their framing is structurally compatible with the industrial-policy calls in America's AI Action Plan but contradicts that plan's preemption-of-state-laws thrust. Federal preemption of state AI laws, as in Executive Order 14365 — Ensuring a National Policy Framework for AI, is incompatible with the multi-jurisdictional regulatory cooperation the essay assumes.
Both authors have Anthropic ties, and the voluntary-commitments-plus-government-collaboration model of the A Framework for AI Development Transparency (Anthropic) maps onto the posture the essay endorses. On financing, the authors acknowledge that capital-market sufficiency is "wonderful for taxpayers" but argue that limits are appearing, which connects to Circular Financing in AI and the Sen. Warren critique of OpenAI.
The essay connects to several wiki pages as a citable source: the CAISI voluntary safety-testing arrangements with frontier labs at CAISI are presented as the most concrete current instance of the bargain; the 50 GW and 12%-of-electricity figures at AI energy demand serve as a citation for energy-as-bottleneck arguments; the October 2024 National Security Memorandum at AI and the military is identified as a framework the authors still endorse, with its incomplete implementation under Trump described as the policy gap; and the talent argument at talent flow is a quantitative articulation of the claim that the US loses its AI lead if Chinese AI students leave.
The authors position the essay, together with Sullivan's "Tech High Ground," as an articulation of how the US AI ecosystem should be governed from a post-Biden Democratic standpoint and as a statement of what a 2028-and-later Democratic AI policy might look like. Both essays are positions rather than factual evidence.
Provenance and confidence
Confidence is medium. The piece is a Foreign Affairs essay treated as Buchanan and Collins's position; Buchanan's Anthropic advisory role is disclosed in the article. The quoted statistics (50 GW, 12%, 70%) should be verified against the underlying NAFSA, Anthropic, and Georgetown analyses before citing.
Relationships
- supports: The Tech High Ground (Jake Sullivan, Foreign Affairs) (companion piece), AI Environmental Impact (energy-bottleneck framing), NIST CAISI (Center for AI Standards and Innovation) (held up as model)
- contradicts: Executive Order 14365 — Ensuring a National Policy Framework for AI (federal preemption posture incompatible with multi-jurisdictional grand-bargain logic)
- related: America's AI Action Plan, A Framework for AI Development Transparency (Anthropic), Circular Financing in AI, Ben Buchanan, Talent Flow China Us (planned)
- instance-of: AI Safety Cases and Frameworks