Washington and the AI sector suddenly are actively engaged in figuring out the federal government’s role in tech use and data center operations. From 2020-2024, one could credibly question whether industry scientific and commercial developments would overwhelm regulation, but 2026 increasingly seems the start of a long, probably deep regulatory engagement.
From an investment perspective, acknowledging a regulatory overhang of AI might prove constructive overall like the benefits of healthcare, commercial flight, ground transportation, and food safety. My observation here of course does not preclude policy missteps – mistakes likely will be a feature of governance not an anomaly. Pacing tech, money, and rules developments become harder every day but that these factors increasingly collide seems obvious.
- Fragmented regulatory oversight of everything enabling AI guarantees policymakers will catch up to capital formation and technology advances of the past decade. Just as data center power use eventually forced sub-federal governmental policy deliberations, power generation and a range of impact pressures fomented First and Second branch ramping involvement.
- The Federal Energy Regulatory Commission today approved guidance that electricity grids show they are working to accommodate large-load interconnections presuming such is feasible and data centers shoulder disproportionate costs of doing so. A rule will be issued 90 days after completion of a subject matter study. One should not be surprised by ensuing litigation regardless of study findings and final rule directions (HERE).
- Trump administration increasing focus on commercial sales and federal government usage of frontier and less capable models seems likely to eventually require Third branch activities. For example, if the federal government can deem one or a handful of data center projects national security critical infrastructure, courts could force the Legislature to determine policy.
- My view remains that federal legislation will be more viable in 2027 rather than in this election year. One benefit of this delay might be congressional appetite for 1H27 action built from 2018-26 hearings, member seminars, and 2026 legislation introduced.
- Policy journey complexities include macroeconomics, human well-being (not just workforce implications) national security, resource allocation, financial stability, and competition guardrails. Policy AImpulse became a section of my Sunday outlook note two months ago, likely two years too late. Nonetheless, the recent edition (HERE) provides links to a handful of policy determinations and recommendations from just this month.
Pending Legislation
A Senate Banking Committee hearing last week is the most recent full committee AI hearing (HERE). Witnesses focused on four broad policy areas: global strategic competition, macroeconomic implications, technological advances, and financial stability (taxpayer bailout protections). Other topics covered in pending House and Senate legislation include intellectual property protections, privacy, environmental impacts, and a long list of issues that broaden out the contours of potential AI regulatory structures.
On economic issues, Will Rinehart of the American Enterprise Institute emphasized task-level gains are real but large and uneven, and they don’t automatically become economy-wide productivity or wage gains. AI helps lower-skilled workers most (the customer-support and legal studies) but can hurt performance on tasks beyond the model’s frontier, and gains depend heavily on training and complementary investment. He invoked the “lightbulb/dynamo/microscope” framing — AI is likely a general-purpose technology whose payoff arrives slowly, like electrification. His recommendations are modest and process-oriented: Congress should focus on scenario planning and better data and should specifically study how AI agents could destabilize financial markets (correlated models amplifying volatility, exploiting regulatory gaps). His appendix on “the economics of compute” argues the data-center boom is real demand, not a hype-driven scheme.
Anthropic sits at the center of practice, research, and policy advocacy of policy-relevant AI activities. Company leaders’ relationship with the White House shapes daily headlines and Administration meetings. For this note, I pull out their policy recommendations as evidence that broad governance rules have eluded lawmakers, a condition I expect to change sooner rather than later. If interested, read the company’s Advanced AI Framework linked in my Sunday note mentioned above. Independent verification of frontier model safety stresses the importance of apolitical reviews. Mr. Amodei, and his lawyers, know very well politics could be a complication difficult to resolve.
A massive cyber event (elections-related or otherwise) or other AI-enabled societal disruption could force legislative action sooner than my 2027 outlook. In this scenario (not a prediction) the November-December so-called lame duck session might need to react to an event creating bottom-up pressures for rulemaking.
Grace Julian, a rising senior at the University of St. Andrews, and 22Vsummer intern contributed meaningfully to this note.
